Home

ClinProTools Manual - Vanderbilt University School of

1. XVa X1 X2 Cap NEP Gen NPC Model Generation Classes 100 100 100 100 Modell GA 1 5 50 true Class 1 D Data Files ClinProTools ClinProTools Test f Data Spiked Data Normal Model2 SVM ne ie ae ie a DAData Files ClinProTools ClinProTools Test a ee U i Data Spiked Data Spiked Figure 4 2 Results of running the basic workflow Model Generation 4 4 3 Basic Workflow Classification The basic workflow Classification can be used to quickly classify test spectra with an existing model This workflow includes selecting the model to use and the spectra to classify data preparation of these spectra according to the settings saved in the model and their classification The classification result is automatically shown in the Classifi cation report and stored as ClinProtClassification number xml file To run the Classification workflow 1 Generate a model as described in the Model Generation workflow Section 4 4 2 or load a previously generated and saved model using the Load Model command from the Model Generation menu or _ toad 4 6 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Getting Started with ClinProTools 2 Select the model to use from the model list Model Name Algorithm Modell GA Model2 SYM Model3 SNN Model4 Qc 3 Load the To Classify 5 5 spectra collection from the ClinProTools Test Data on the CD using the
2. 1 2 Click Programs 3 Click Bruker Daltonics 4 Click Administration 5 Click LicenseManager to open the Bruker Daltonics LicenseManager dialog fam Bruker Daltonics gt f Administration gt LicenseManager 6 Enter the ClinProTools license key in New license key 7 Click Add The button is enabled after entering a correct license key If the key is valid the license is added to Existing licenses Figure 2 1 8 Repeat steps 6 and 7 for the Support Vector Machine license key if available 9 Click Close f for temporary New license key licenses only Existing licenses License Key Valid Until Product Name unlimited ClinProTools 2 2 unlimited Support Vector Machine 1 0 Figure 2 1 Bruker Daltonics LicenseManager dialog with the ClinProTools 2 2 and Support Vector Machine 1 0 licenses present 2 4 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Installing and Licensing ClinProTools 2 5 Uninstalling ClinProTools There is no need for uninstalling previous ClinProTools versions when a new ClinProTools upgrade version should be installed on your system Nevertheless if you want to uninstall ClinProTools from your system proceed as follows To uninstall ClinProTools cio MED Click Settings Click Control Panel Double click Add Remove Program Select the ClinProTools 2 2 software from the list of installed programs Click Remove Confirm
3. D Data Files ClinProTools ClnProTools 7 Test Data Spiked Data To Classify 5 5 true 2 0 05 1 95 0 215 0 128 0 233 O_N17_1SLin_S fid E Figure 8 9 Classification report for spectra when a QC model was used which is based on single spectra peak calculation and without information from the ClinProtRobot section 8 10 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Reporting Data The following data is displayed for each spectrum Column Description Index Order of spectra loading Name Path and name of the spectrum Classified Whether or not the algorithm was able to classify the spectrum Class Estimated class Class N To which extent the spectra belongs to class N gt 1 good 1 average lt 1 bad for QC only State Spectrum state information Score Recogniton score of the classified spectrum matched against the overall average peak list for single spectra peak picking only Score N Recognition score of the classified spectrum matched against the per class average peak list for single spectra peak picking only The following items are only generated for spectra with information from ClinProtRobot Column Description Sample Name Sample name Sample ID One to one ID Sample Group Group membership like disease or normal Sample Type Kind of the sample Patient Patient name Comment Arbitrary comment Source Source plate name and position Processin
4. Prototype region Prototype regions representing a set Da each representing of data points of ee a set of data points the green class p i A ofthe blue classi y c S al 5 E kd mension A prototype x ofthe green class oe located in the center of a green cluster Aprototype x of the blue class located inthe center of a blue clusterf Figure 6 3 Supervised Neural Network algorithm Determination of class prototypes The SNN has to learn the characteristic of the two classes in a way that new data points can be classified to one of the two classes To do this the SNN tries to determine a set of prototypes depicted as x green class and x blue class Multiple prototypes for one class are typical Thereby the distance between separating boundaries the polygons in the figure of consecutive classes e g green blue in the figure should be as large as possible for the SNN this is formalized in an optimization problem solved by a gradient descent on a cost function similar as within a neural network approach The prototypes x and x represent a set of data points from the original data set This means that all data points which are closer to a prototype e g from the prototype set of the green class than to a prototype II belong to prototype Thereby it is claimed that this prototype is characteristic for these data points If the prototype is ClinProTools User Ma
5. Width Da Enter the smoothing width in Dalton Cycles Enter the number of smoothing iterations In Data Reduction define the parameters for the data reduction filter Section 6 1 3 1 Enable Check this option if the data reduction filter should be enabled Setting this option allows speeding up calculations and reducing the memory consumption especially for very large data sets However best classification results are expected without data reduction Note The data reduction is applied prior to any other data processing and influences all subsequent results Factor Enter the data reduction factor This sets the number of consecutive data points in a set that are to be replaced by the average of these points Typically the value should be chosen between 2 double reduction and 10 10 fold reducetion The greater the factor is chosen the more features will be smoothed out As a consequence e g shoulder peaks may no longer be resolved ClinProTools User Manual Version 2 2 9 33 Reference Part ClinProTools Menus Bruker Daltonik GmbH In Null Spectra Exclusion define the parameters for the null spectra exclusion filter Section 6 1 3 2 Enable Check this option if the null spectra exclusion filter should be enabled to sort out spectra with a very low intensity In general this option should be disabled only in the case of third party spectra In Noise Spectra Exclusion define the parameters for the noise s
6. Basics Bruker Daltonik GmbH 6 2 Model Generation and Validation In ClinProTools models are generated which function as classifiers The aim of model generation is to describe the spectra of the model generation classes in such a way that new spectra can be classified afterwards ClinProTools supports four kinds of algorithms for generating classification models In general the classification results can be improved by a meaningful restriction of the number of peaks This can be done during the data preparation due to a selection based on the signal to noise ratio or other criteria Limit Peak Number parameter in the Settings Peak Calculation dialog Section 9 1 4 2 Another possibility is to select the peaks according the Sort Mode in the Settings Peak Selection dialog Section 9 1 5 1 which allows using only the peaks with the probably highest class separation capability under a univariate view on the data The following sections provide information on ClinProTools classification algorithms k nearest neighbor classification cross validation and external validation 6 2 1 Classification Algorithms ClinProTools supports four kinds of algorithms for generating classification models All these algorithms are different in their methodology and have advantages and draw backs Figure 6 1 illustrates the characteristics of the four classification algorithms e Genetic Algorithm GA This algorithm mimics evolution in nature and
7. D Data Files ClinProTools ClinProTools Test Data Spiked Data To Classify 5 5 0 L15_1SLin Nid mus 2 D Data Files ClinProTools ClinProTools Test Data Spiked Data To Classify 5 5 0_L17 _15Lin Nid Eue 3 D Data Files ClinProTools ClinProTools Test Data Spiked Data To Classify 5 5 O_L19_1SLin_N fid true jl Figure 4 3 Results of running the basic workflow Classification ClinProTools User Manual Version 2 2 4 7 Getting Started with ClinProTools Bruker Daltonik GmbH 4 5 Closing ClinProTools You can close ClinProTools when you have finished your current session To exit ClinProTools you have to answer a confirmation request To close ClinProTools 1 From the File menu select Exit or click the application s 2 Answer the confirmation request to close ClinProTools 4 8 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH ClinProTools User Interface 5 CLINPROTOOLS USER INTERFACE On starting ClinProTools the ClinProTools window opens Herein all processing operations ClinProTools supports are started and most of the results displayed Only results of a Principal Component Analysis PCA or an unsupervised hierarchical clustering display in separate windows the PCA windows and the Dendrogram window which originate from the external MATLAB software tool integrated in ClinProTools 5 1 ClinProTools Window The ClinProTools window Figure 5 1 consists of four vie
8. Spectra grouping parser The spectra grouping parser determines the spectra group membership by analyzing the spectra paths names It works as follows The list of spectra paths of a spectra collection is processed sequentially If consecutive paths have the same group folder they are considered to be belonging to one group In reverse order the group folder is the fourth subfolder of the fid file path of a spectrum E g in the case of F F1 0_D9 1 1SLin fid F1 is the assumed group folder The first sub folders name must be one of 1SLin 1Lin 1Ref 1slin 1lin or 1ref 1SLin in the example The second must be a number from 1 to 9 1 in the example The third subfolder must contain an underscore 0_D9 in the example If these conditions are not given the spectrum will not be considered belonging to a group The following example will be parsed as two groups F1 and F2 with four spectra each F F 1 0_D1 1 1S Lin fid F F 1 0_D2 1 1S Lin fid F F 1 0_D3 1 1S Lin fid F F 1 0_D4 1 1S Lin fid F F2 0_D5 1 1S Lin fid F F2 0_D6 1 1SLin fid F F2 0_D7 1 1SLin fid F F2 0_D8 1 1SLin fid ClinProTools User Manual Version 2 2 6 7 Basics Bruker Daltonik GmbH Note To make sure that the spectra are parsed correctly verify the spectra grouping performed by the parser The grouping is displayed in the Groups column in the Spectra List report Section 8 1 1 1 In the Gel View gr
9. e Graphic resolution 1024x768 pixels 256 colors or better optimum 1280x1024 with true colors e CD ROM DVD drive only for installation e Microsoft NET Platform will be installed by Setup if not found on computer ClinProTools User Manual Version 2 2 2 1 Installing and Licensing ClinProTools Bruker Daltonik GmbH 2 2 Installing ClinProTools The ClinProTools software is installed from the ClinProTools installation CD If the MATLAB Component Runtime application is not available on your system you will be prompted installing it prior to the installation of ClinProTools Installation notes e The program should be installed by a user with administrator rights it is not sufficient to install it as a normal user with Run as administrator e During setup the installation of the Net framework must be affirmed e Internet Explorer 6 0 is required and for loading the XML files with style sheets Excel 2002 or higher Make sure that the Excel security settings extras options securi ties macro security are set to low During the setup the MATLAB Component Runtime will be installed Please check All Users during MATLAB Component Runtime setup to ensure proper access for all users e Sometimes empty tables occur while displaying XML with style sheets in the Internet Explorer This is due to an out of date XML parser registered Microsoft provides a Replace Mode Tool called Xmlinst exe which sets the applicatio
10. 9 1 3 6 10 Spectra View gt Outliers for Box amp Whiskers Command The Outliers for Box amp Whiskers command shows hides the outliers for the per class box amp whiskers plots in the Spectra View and also in the Single Peak Variance View when the Spectra View gt Box amp Whiskers command is active This toggles the box amp whiskers plots between the standard box plot command not active and the modified box plot command active Figure 9 14 The modified box plot differs from the standard box plot with respect to the meaning of the displayed whiskers and the additional display of outliers which are measured val ues that do not fall inside the whiskers In the modified box plot the end mark of the top whisker denotes the value that is calculated by adding 1 5 times the interquartile range range between 25 and 75 quartile to the largest measured value which is smaller than or equals the 75 quartile The end mark of the bottom whisker denotes the value that is calculated by subtracting 1 5 times the interquartile range from the smallest measured value which is larger than or equals the 25 quartile Thus ca 95 of all measured values are inside the whiskers if the whisker length is 1 5 times the inter quartile range All measured values that are larger or smaller respectively than the end marks of the whiskers are indicated as outliers Outliers are denoted in the modi fied box plot by symbols which correspond to the symbols u
11. Reference Part ClinProTools Context Menus 9 2 ClinProTools Context Menus 9 2 1 Spectra View Context Menu The Spectra View context menu offers the following commands Command Coordinates Grid Scaling Auto Scaling Zooming Undo Zoom Redo Zoom Distance Display Mode Background Color View Spectrum Info Exclude Include Spectrum Exclude Include Peak Force Peak into Model Show Spectrum Add Peak Remove Peak Edit Peak ROC Curve for Peak Variance for Peak Correlation List for Peak Used to Show Hide the display of cursor coordinates in the status bar Show Hide the grid in the view Pop up scaling commands for the view Activate Deactivate auto scaling in the view Activate Deactivate the zoom in mode in the view Same as Undo Zoom command from View menu Same as Redo Zoom command from View menu Switch the view to distance measurement mode Pop up display modes for the view Define the background color of the display region of views Show the spectrum info for the selected spectrum Same as Exclude Include Spectrum command from Edit menu Exclude Include the selected peak in model generation Force the selected peak into the next generated model If the peak distribution or the box and whisker plot with outliers is activated show in the Spectra View the spectrum that corresponds to the right clicked data point in peak distribution Add a new peak to the peak list Remove the selec
12. and the sample set is the given set of spectra A correlation matrix resp correlation list is set up as the result of correlation analysis Crossover Crossover means the combination of two randomly selected individuals to produce two new individuals by interchanging parts during GA Cross validation This type of Validation should be used for automatic validation during model genera tion During cross validation a small part of all spectra is left out in model generation and cluster analysis These spectra are then classified and the number of correct and wrong class predictions is determined This procedure is repeated several times and the correct and wrong class predictions are accumulated for each class External validation This type of Validation uses a separate test set which has not been used for generating the Classification model For all spectra of the test set the true class mem A 6 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Appendix bership has to be known During validation all spectra of the test set are classified The predicted class memberships are compared to the true class memberships and Sensitivity and Specificity can be calculated Generation During one generation of a Genetic Algorithm a new Population is created Genetic Algorithm GA The Genetic Algorithm is a stochastic search algorithm which mimics evolution in nature It is used for the optimization of an objective functi
13. clinprotools 2 2 User Manual clinprotools 2 Bruker Daltonics Bruker Daltonik GmbH Copyright Copyright 2007 Bruker Daltonik GmbH All Rights Reserved Reproduction adaptation or translation without prior written permission is prohibited except as allowed under the copyright laws Document History ClinProTools User Manual Version 2 2 November 2007 Part 249619 First edition June 2004 Printed in Germany Warranty The information contained in this document is subject to change without notice Bruker Daltonik GmbH makes no warranty of any kind with regard to this material including but not limited to the implied warranties of merchantability and fitness for a particular purpose Bruker Daltonik GmbH shall not be liable for errors contained herein or for incidental or consequential damages in connection with the furnishing performance or use of this material Bruker Daltonik GmbH assumes no responsibility for the use or reliability of its software on equipment that is not furnished by Bruker Daltonik GmbH Bruker Daltonik GmbH Fahrenheitstrasse 4 28359 Bremen Germany Phone 49 421 2205 432 FAX 49 421 2205 106 E mail clinprot support bdal de Internet http www bdal de ii ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Contents Contents 4 PREFACE icici aE EREE EEEE TERE EOAR VEA 1 1 2 INSTALLING AND LICENSING CLINPROTOOLS cc cccsssecsessessese
14. 6 32 9 24 Peak Statistics View popup command 9 24 ROC Curve command i i 9 24 Be e ee spectra 6 4 7 6 ROC Curve for Peak command 9 77 a a ROC Curve View 5 7 Plots menu MATLAB oe ROC Curve View context menu 9 69 Principal Component Analysis 6 34 Rotate command MATLAB 9 84 Print command 9 5 Print Preview command 9 5 Ss Print Setup command 9 6 Printing Sample preparation for clinical Data plotting view 8 12 proteomics 3 1 Report 8 12 Save Class Paths command 9 5 P values 6 30 6 38 Save Classification command 9 56 Save Model As command 9 78 ClinProTools User Manual Version 2 2 I 5 Index Bruker Daltonik GmbH Save Settings Data Preparation command Save Settings Model Generation command Saving Classification result Data preparation settings Model Model generation settings Report Spectra import XML file Savitsky Golay smoothing Scaling command Scores PCA Scores plot PCA Select Peaks command Sensitivity Setting Cross validation parameters Peak calculation parameters Peak selection parameters 7 2 Spectra preparation parameters Settings Peak Calculation command Settings Peak Selection command Settings Spectra Preparation command Settings Statistic command Shortcut reference Show Classification command Show Error command Show Model command Show Spectrum command Showing Classification result Model Model list Peak list Report Similarity selection filter Single PCA plot window PCA Single Peak Variance comman
15. 95 confidence interval 9 27 Cancel command 9 4 9 50 95 Confidence Interval command 9 27 CART peak list export format A 13 Class Names command 9 21 A Class opening 7 5 Abbreviations A 11 E z e About ClinProTools command 9 66 Closing 7 18 Add Peak command 9 70 Modes 6 25 Adduct Polymer spectra exclusion Running 7 17 fier 9510 Fat Saving result 7 17 All Single Spectra command 9 14 Selecting spectra 7 17 Altering data plotting views 5 11 noes 7 18 EA E test ae Classification algorithms 6 12 ASCII i i A 12 Classification menu 9 54 ae ee Baas Classification report 7 17 7 18 8 10 Nit a d f 9 71 Classification standard workflow 4 6 Pea e comman 512 Classify command 9 54 uto sca Ing Classifying spectra 4 6 6 25 7 16 Average amp StdDev command 9 16 Clear All command 9 50 Average peak list calculation 6 4 7 7 ClinProTools Average Peak List Calculation Basic workflows 4 3 command a Clearing temporary XML files 4 2 Average peak list calculation workflow 7 7 Closin 4 8 Average Spectra command 9 15 rahe shutdown 10 1 Average spectra display 9 15 File location 4 2 Average with standard deviation 5 8 9 16 General settings 4 2 ClinProTools User Manual Version 2 2 l 1 Index Bruker Daltonik GmbH Installing Licensing Reports Starting Supporting more than 2 GB RAM System requirements Uninstalling User interface ClinProTools window ClinProTools XML files Clearing temporary files Creating Close All command Close Classificat
16. Figure 9 62 Only included peaks are passed on to the algorithms In the Peak Statistic report Section 8 1 1 2 excluded peaks are indicated by a entry in the S state column first column arb u D Data Files ClinProTools ClinProTools Test Data EDTA Run0hiSample0_E10_1SLin fid 120 i i Bi 100 80 60 40 20 E 1350 1400 1450 1500 1550 miz Figure 9 62 Display of included blue and excluded gray peaks in the Spectra View 9 76 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Reference Part ClinProTools Context Menus 9 2 9 14 Force Peak N into Model Command The Force Peak n into Model command forces the selected peak into the model to be generated Peaks can be forced after running the peak calculation workflow A forced peak is marked by a green integration region before as well as after model generation Forcing a peak into a model can be canceled by selecting the command for the respective peak again 9 2 9 15 Grid Command Command The Grid command shows hides the grid in a data plotting view This command applies to the selected view only The grid properties cannot be changed 9 2 9 16 Remove Model Command The Remove Model command removes the selected model from the Model List View Note Please remember that models are not automatically saved in ClinProTools Thus if you have calculated a new model you should first consider if you want to save it Section 9 2 9 19 before removing it 9 2 9
17. For this select the folder with the spectra you want to load as a class ClinProTools loads all spectra in a folder and its subfolders as one class and prepares them Add model parameter sets to the Model List View using the New Model command from the Model Generation menu or Seita For this select the classification algorithm GA SVM SNN or QC click OK in the appearing algorithm specific settings dialog to use the defaults and enter a model name Repeat this procedure for each of the four algorithms Start model calculation using the Calculate command from the Model Calculation menu or _Calculate This runs data preparation spectra recalibration spectra averaging and peak calculation on the loaded spectra and after that generates a model for each added model parameter set The generated models are entered in the model list To view the parameters of all models in the model list use the Model List com mand from the Reports menu or 0 List Section 8 1 1 5 Figure 4 2 To view a model select it from the list and use the Show Model command from the Show This opens the Model List report Model List View context menu or Model Name Algorithm Model2 SVM Model3 SNN Model4 Qc This opens the Model report Section 8 1 1 6 which lists all parameters of the selected model Figure 4 2 The Spectra View Section 5 1 1 shows the peaks that are incorporated in the current model now having red integration region
18. Mass Tolerance Da fi fa 21 980000 1 000000 Figure 9 31 Adduct Polymer Property dialog for adding a new left or editing an exist ing adduct polymer right Remove removes the selected adduct polymer ClinProTools User Manual Version 2 2 9 35 Reference Part ClinProTools Menus Bruker Daltonik GmbH In Spectra Grouping define the parameters for spectra grouping Section 6 1 2 and the similarity selection filter Section 6 1 3 2 Support Spectra Grouping Check this option if multiple spots of one sample should be treated as a group spectra grouping Note This option is suitable only for automatically created spectra by the current ClinProtRobot with the corresponding software If the option is enabled while using a different folder structure the parser might by coincidence detect not existing groups which will lead to calculation errors If switched on manually copied spectra may also be parsed as a group if it has the same folder structure as generated by the ClinProtRobot Enable Similarity Selection Check this option if the similarity selection filter should be enabled to detect the most suitable spectrum of a spectra group not suitable spectra will be excluded In Recalibration define the parameters for recalibration Section 6 1 1 3 and the spectra quality filter Section 6 1 3 2 Enable Check this option if spectra should be recalibrated ppm Maximal Peak Shift Enter the maximal m
19. The settings are automatically stored in the SettingsDataPrepara tion xml file which is loaded when ClinProTools is started and is updated on each settings change To keep special settings you can save them in an XML file Although only used in model generation the Settings Peak Selection parameters can also be specified in the context of defining the data preparation settings since initial peak selection is a part of the peak calculation workflow The settings are automatically stored in the SettingsModelGeneration xmI file 7 1 1 1 Setting the Spectra Preparation Parameters The Settings Spectra Preparation dialog Section 9 1 4 1 defines the settings for preparing spectra Most of the parameters apply to filters that modify or select spectra during spectra loading Some parameters also apply to peak picking for spectra recalibration as well as to spectra recalibration itself and averaging You can use the default parameters or specify own settings suitable for your data Alternatively you can ClinProTools User Manual Version 2 2 7 1 Workflows in Detail Bruker Daltonik GmbH load a data preparation settings file or reset the current settings to the default values Section 7 1 1 4 Note If these parameters are changed after spectra loading or even processing the views may become cleared to prevent the spectra from further processing and a message will inform you on how to proceed To set the spectra preparation parameters 1
20. Version 2 2 Bruker Daltonik GmbH Basics 6 2 2 K Nearest Neighbor Classification The k nearest neighbor k NN classifier algorithm is used within the GA and SVM to obtain the final classification It just uses the distances between points in the n dimensional space Remember that each point corresponds to a spectrum The coordi nates of the point are made up of the peak areas of the spectrum The peak selection is derived from the current GA peak combination or the final SVM peak ranking solution The idea of k NN classifiers is to look at the k nearest neighbors and their spectra class membership For details on numerical analysis of the k NN principle we refer to T Hastie R Tibshirani and J Friedman The Elements of Statistical Learning Springer 2002 Workflow The workflow of the k NN classification is as follows 1 The distances between all data points spectra are calculated 2 The k nearest neighbors for each point are determined 3 Each point is classified according to the class membership of the neighboring points Figure 6 4 4 The separation value is calculated which indicates how good the data could be separated and classified with the current parameter of k NN and the given peak selection Legend class 1 o class 2 Figure 6 4 Classification of points A and B using three nearest neighbors Parameterization The Number of Neighbors parameter in the Settings Genetic Algorithm Support Vect
21. different mass ranges and it is recommended for inexperi enced users to start working with one of them and to adapt the method to their special approaches In the following two model methods created on Autoflex and Ultraflex MALDI TOF MS respectively for generating clinical proteomics profiles are described Table 3 1 These parameters are for information only and are to be regarded only as a guide they should not simply be copied to the user s mass spectrometers as each machine can differ They represent first easy to use starting parameters for the profiling user Usually only slight changes are necessary to create an individual method according to each machine Gray highlighted values in the table represent values that do not signi 3 2 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Data Acquisition for Clinical Proteomics ficantly change between different mass spectrometers and can generally be defined as fixed Table 3 1 Recommended method parameters linear mode for measuring clinical proteomics profile spectra Note Please note that parameter sets have to be optimized for differ ent proteomic samples and for every instrument Parameters Autoflex 1 10 kDa Ultraflex 1 10 kDa N2 pressure approx 1700 2000 mbar Laser individually adjustable individually adjustable it is recommended to it is recommended to shoot shoot approx 15 shots approx 15 shots with with higher laser power highe
22. lt 0 000001 862 52 95 0 65 7 25 Tse Teh x 10 1046 39 32 48 lt 0 000001 lt 0 000001 lt 0 000001 i 2 82 j 35 3 0 55 6 03 i 19 45 17 09 X 5 933 49 6 28 lt 0 000001 lt 0 000001 0 909 24 43 18 15 299 2 95 12 25 16 26 6 940 41 3 83 0 00000176 0 0000012 0 844 15 25 11 42 1 99 1 83 13 02 16 06 X 13 1149 46 3 05 0 0000122 0 00000414 0 844 16 16 13 11 1 55 1 84 9 58 14 Mi nS 1120 41 9 049 A ANNAN A ANNNANHAEO no 12901 1n 92 1Aa JiS ing 1 5A Figure 8 3 Peak Statistic report section 8 4 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Reporting Data 8 1 1 3 Correlation Matrix Report The Correlation Matrix report ClinProtCorrelationMatrix xml Figure 8 4 is created and shown using the Correlation Matrix command from the Reports menu The correla tion analysis Section 6 4 2 1 is based on the settings defined in the Correlation Matrix dialog The report shows the correlation matrix that lists all peak pairs with their calculated correlation coefficient cc ranging from 1 to 1 In addition a color code ranging from red cc 1 to blue cc 1 is used to highlight different cc ranges in the matrix Th is allows quickly detecting highly correlated peak pairs Whether correlation was calcu lated over all classes or a specified one is reported above the table as well as the used correlation algorithm standard or Kendall s tau the s
23. with ClinProTools we are focused with the multiple measurement mm problem Sec tion 6 4 3 3 If mm are used they must be processed as mm This has to be specified by the user in the spectra preparation settings ClinProTools then automatically processes the mm in the correct form sample by sample If mm are processed without the corresponding option set each spectrum is considered as an individual sample This would lead to invalid cross validation statistic results Unequal class sizes Unequal class sizes are very common in clinical research especially by considering cancer and control classes This has some effects which should be kept in mind For the data preparation the peaks are picked on an average spectrum In the next step of the statistic calculation very small sample sizes for one class may give a bias in the test procedure In addition the classification procedures are more or less affected by different sample sizes For the GA the optimization could be dominated by e g one large class The QC relies on some statistical measurements hence the QC may be affected by unequal sample sizes similar as the statistic calculation The SVM looks for extreme boundary points to determine the hyperplanes hence it is less effected by unequal sample sizes as long as the boundaries have a clear definition However in ClinProTools User Manual Version 2 2 6 37 Basics Bruker Daltonik GmbH the cross validation it may happen that e
24. 1 should be used if the number of samples is very small For a larger number of samples per class k gt 1 is recommended Advanced gt gt Advanced lt lt Expands Contracts the dialog to display hide the advanced GA parameters In Initial Number of Peak Combinations define how to determine the initial number of peak combinations within the population Automatic Detection Check this option if the initial number of peak combinations should be determined automatically To automatically determine the number of peak combinations npc the following heuristic formula is used NPC 100 NumberOfPickedPeaks x 20 MaximalNumberOfPeaksInModel 1 Number of Peak Combinations Enter the initial number of peak combinations if Automatic Detection is not set Mutation Rate Enter the mutation rate which is the likelihood of a mutation In ClinProTools a muta tion is the random exchange of a peak within a peak combination by a randomly selected new one The values can range from 0 0 no mutation occurs to 1 0 all peak combinations are mutated in each generation Crossover Rate Enter the crossover rate which is the likelihood of a crossover between peak combina tions The values can range from 0 0 no crossovers to 1 0 all peak combinations in each generation are used in crossover and are replaced by their children Use Varying Random Seed Since the GA employs random numbers for selection crossover and mutation it is possib
25. 1 5 should by performed which automatically determines the best number of peaks to be integrated in the model by an internal iteration The search for the number of best peaks is restricted to maximal 25 peaks in a model Number of Peaks If Automatic Detection 1 25 Peaks is not set enter the number of peaks that must be integrated in the model OK Opens the Model Name dialog Section 9 1 5 2 5 to specify a name for the model 9 1 5 2 3 Settings Supervised Neural Network Dialog The Settings Supervised Neural Network dialog Figure 9 38 defines the basic and advanced parameters for the SNN The SNN automatically uses automatic peak detection Section 6 2 1 5 to determine the best number of peaks to be integrated in ClinProTools User Manual Version 2 2 9 47 Reference Part ClinProTools Menus Bruker Daltonik GmbH the model maximum is 25 peaks The settings are stored as described for the GA settings Section 9 1 5 2 1 Settings Supervised Neural Network The number of peaks in the model will be detected automatically 1 25 peaks ox ca Cor 10 x 100 Upper Limit of Cycles J Automatic Detection of Prototype Number Number of Prototypes Figure 9 38 Settings Supervised Neural Network dialog default setting Advanced gt gt Advanced lt lt Expands Contracts the dialog to display hide the advanced SNN parameters Upper Limit of Cycles Enter a value multiplied by 100 for the upper limit of cycles
26. 2 Peak Statistics View gt 2D Options gt 95 Confidence Interval Command The 95 Confidence Interval command toggles the 2D Peak Distribution View between displaying the calculated 95 confidence interval command is active Figure 9 25 or standard deviation command is inactive Figure 9 26 for each class as an ellipse The 95 confidence interval is the standard deviation weighted by the recipro cal number of data points The ellipses are displayed according to the classes color coding For information about the confidence interval we refer to J M Chambers and T J Hastie Statistical Models in S Wadsworth amp Brooks Cole 1992 ClinProTools User Manual Version 2 2 9 27 Reference Part ClinProTools Menus Bruker Daltonik GmbH Pk 9 1467 Da 0 20 40 60 80 100 Pk 16 1898 Da Figure 9 25 Display of 95 confidence interval Pk 9 1467 Da 0 20 40 60 80 Pk 16 1898 Da Figure 9 26 Display of standard deviation 9 1 3 8 5 3 Peak Statistics View gt 2D Options gt Current Spectrum Marker Command The Current Spectrum Marker command shows hides in the 2D Peak Distribution View the marking of that data point that corresponds to the current spectrum in the Spectra View Figure 9 27 The respective data point is marked by bold display 9 28 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Reference Part ClinProTools Menus Pk 9 1467 Da 5 10 15 20 25 Pk 16 1898 Da Figur
27. 5 2 1 If you want to set up the Influence plot for a different number of PCs again select the Influence command and enter the desired number of PCs 7 5 2 3 Variance Plot The Variance plot Figure 5 16 can be displayed via the Variance command in the Plots menu It displays the explained variance in percent contributed by the single given PCs The blue curve starting from the first bar demonstrates the accumulated variance from PC to PC The number of PCs concerned in the plot depends on in the investigated data Basically the plot displays as many PCs as are needed to explain at least 95 of the variance within the data set but it is limited to displaying ten PCs at most This may result in the sum of explained variance curve not reaching the 95 mark in each case 7 6 Performing Unsupervised Clustering To get more information about the variability within model generation classes and thus the homogeneity heterogeneity of a spectra set an unsupervised hierarchical cluster ing can be carried out within ClinProTools In the context of unsupervised clustering the separation into classes is ignored i e all data is treated as one group It is also possible to apply unsupervised clustering to a single loaded class only e g to detect subgroups or outliers within the model generation class Unsupervised clustering can be performed on grouped spectra too The unsupervised clustering is carried out by an external MATLAB software tool which is st
28. 8 1 Displaying details of a report here for the item S 8 1 1 ClinProTools Report Types 8 1 1 1 Spectra List Report The Spectra List report ClinProtSpectra xml Figure 8 2 is created and shown using the Spectra List command from the Reports menu This report lists all loaded spectra with corresponding data grouped according to their class membership The following data is displayed for each spectrum Column Description Name Path and name of the spectrum State Inclusion exclusion state of the spectrum Excluded spectra are marked by an Excluded entry In case of exclusion by filters the filter is given e g Excluded Noise The rows of excluded spectra are colored according to the reason of exclusion the color code is the same as used in the Gel View Section 9 1 3 7 3 Sample Name Sample name Mean Intensity Average intensity before TIC normalization Laser Shots Number of shots Spectrum ID ID of the spectrum Groups Grouping of spectra available when the Support Spectra Grouping option in the Settings Spectra Preparation dialog is enabled 8 2 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Reporting Data gt ClinProt Spectra List ClinProTools Version 2 2 build 28 B R U K E R Class 1 00h LXK Sample Mean Laser pan pee Name Intensity Shots Spectrom ID D Data Files ClinProTools ClinProTools Test Sample 92 09 120 84cc2b0c 5755 4a91 Data EDTA Run 00h Sample O_E10_1SL
29. 9 2 9 10 5 1 2 Gel Stack View The Gel Stack View consists of two views Gel View and Stack View You can toogle between the views using the Display Type gt Gel View and gt Stack View commands from the view s context menu The Gel View is displayed by default ClinProTools User Manual Version 2 2 5 3 ClinProTools User Interface Bruker Daltonik GmbH 5 1 2 1 Gel View The Gel View Figure 5 3 displays all spectra of the loaded classes arranged in a pseudo gel like look The x axis records the m z value The left y axis displays the running spectrum number originating from subsequent spectra loading The peak inten sity is expressed by a color code The color bar and the right y axis indicate the relation between the color a peak is displayed with and the peak intensity in arbitrary units Various color modes are available Section 9 2 9 6 arb u Oo D Data Files ClinProTools ClinProTools Test Data EDTA RuniTo Classify 140 120 100 80 60 40 ial D Data Files ClinProTools ClinProTools Test DataiEDTA Run0h lt 20 0 2000 4000 6000 8000 miz Figure 5 3 Gel View showing the spectra of five model generation classes red green blue ocher violet and a spectra collection to be classified black using a linear gray scale for intensity display The spectra of the first loaded class class 1 are displayed at the bottom of the view the spectra of the second loaded class class 2 above the class 1 spectra
30. Classify command from the Classification menu or Dessi This prepares the spectra according to the parameter settings saved with in the current model and classifies the spectra 4 View the Classification report Section 8 1 1 8 which opens automatically and lists the classification results Figure 4 3 The Gel View Section 5 1 2 1 and Spectra View Section 5 1 1 now also display the spectra of the classified spectra collection Figure 4 3 M ClinProTools Class 1 Normal Class 2 Spiked To Classify To Classify 5 5 e Edit Yie Jata Preparation Model Generation Classification Statistical Analysis Reports Co Help woe Slr Siac als A E ce ENE a am New ance Show Classify Load Clear All Model List Save Validate Model Name Algorithm Cross Validation Recognition Capability State GA 100 00 100 00 Calculated SVM 100 00 100 00 Calculated SNN 100 00 100 00 Calculated ac 100 00 100 00 Calculated ak File Edit Yiew Favorites Tools Help gt x a O P Search She Favorites Az B G ea LJ rel 33 Address E C BDAL ClinProTools_2_2 Files ClinProtClassification0001 xml ClinProt Classification Spectra Collection D Data Files ClinProTools ClinProTools Test Data Spiked B R U K E R Path Data To Classify 5 5 Model Name Modell Date Time 2007 05 16T09 48 56 535 02 00 ClinProTools For Help press F1 Version 2 2 build 38 Index Name Classified Class State
31. From the Data Preparation menu select Settings Spectra Preparation 2 In the Settings Spectra Preparation dialog specify the parameters as desired and click OK 3 If the views become cleared quit the message and do the required action 7 1 1 2 Setting the Peak Calculation Parameters The Settings Peak Calculation dialog Section 9 1 4 2 defines the settings for peak picking on either the total average spectrum or the single spectra and peak calculation in the individual spectra You can use the default parameters or specify own settings suitable for your data Alternatively you can load a data preparation settings file or reset the current settings to the default values Section 7 1 1 4 Note If the parameters are changed after spectra processing the views may become cleared to prevent the spectra from further processing and a message will inform you on how to proceed To set the peak calculation parameters 1 From the Data Preparation menu select Settings Peak Calculation 2 In the Settings Peak Calculation dialog specify the parameters as desired and click OK 3 If the views become cleared quit the message and do the required action 7 1 1 3 Setting the Peak Selection Parameters Although the peak selection becomes effective only in model generation it is part of the peak calculation workflow and thus its settings are specified in the context of defining the data preparation settings However you can change the curre
32. Genetic Algorithm 5 Maximal Number of Peaks in Model 50 Maximal Number of Generations KNN Classification 3 X Number of Neighbors Cancel Help Advanced lt lt Initial Number of Peak Combinations V Automatic Detection Number of Peak Combinations 0 20 Mutation Rate 0 50 Crossover Rate I Use Varying Random Seed Figure 9 36 Settings Genetic Algorithm dialog default setting Maximal number of Peaks in Model Enter the maximal number of peaks included in the model Maximal Number of Generations Enter the maximal number of generations iterations for the algorithm to run Most of the time this number will not be reached as the stop criteria will halt calculation when no better peak combination is found for a number of iterations In KNN Classification define how to perform k nearest neighbor classification Sec tion 6 2 2 ClinProTools User Manual Version 2 2 9 45 Reference Part ClinProTools Menus Bruker Daltonik GmbH Number of Neighbors Enter the number of neighbors k to be used Per default k can be set only to the odd values 1 3 5 and 7 which has been found to perform reasonable well on different data sets The odd value ensures that in general a classification is obtained using k NN unclassified may still happen for e g three classes and k 3 where two neighbors belong to different classes and that the solution is sufficiently stable The case of one neighbor k
33. Menus arb u ols Test Data EDTA RunvO0hiSample_E10_1SLin fid 2015 2020 2025 2030 miz Figure 9 11 Average with standard deviation bars marking the peak area intensity averages of five classes 9 1 3 6 8 Spectra View gt Peak Distribution Command The Peak Distribution command shows hides the 1D distribution of the peak areas intensities in the Spectra View Figure 9 12 The 1D peak distribution plots the areas intensities of the respective peak in the single spectra of the loaded classes as separate values Like in the 2D peak distribution values of peaks from different classes are displayed with different predefined symbols e g cross circle that are colored according to the respective class color The plot is drawn on a unique scale indepen dent of the peak intensity scale The 1D peak distribution is hidden by default arb u ols Test Data EDTA RunWO0hiSample_E10_1SLin fid 25 20 2015 2020 2025 2030 mz Figure 9 12 1D peak distribution plotting the single areas intensities for the respective peak in the spectra of five classes ClinProTools User Manual Version 2 2 9 17 Reference Part ClinProTools Menus Bruker Daltonik GmbH Shortcut oo Button 9 1 3 6 9 Spectra View gt Box amp Whiskers Command The Box amp Whiskers command shows hides the per class box amp whiskers plots for the peak area intensity in the Spectra View Figure 9 13 In this standard box plot the top and bottom end
34. PCA is to reduce the dimensionality of a data set while simultaneously retaining the information present in the data In data sets with many groups of variables variables often show similar behavior and contain redundant information In the case of mass spectra the variables are represented by the intensity at defined masses According to the resolution the number of these variables can be very high The PCA reduces the number of dependent variables contained within the spectra set via replacing groups of variables by a single new vari able By this a set of new variables so called principal components will be generated Each principal component PC is a linear combination of the original variables All principal components are orthogonal to each other so there is no redundant informa tion In many cases depending on the complexity of the data set only few PCs com pared to the large number of original variables contain most of the variance The full set of PCs is as large as the original set of variables nevertheless only the first PCs are of interest mostly higher PCs contain very detailed spectra information and the highest PCs contain spectra noise Figure 6 8 describes the transformation of a data set to PCs in a simplified graphic Actually each sample spectrum can be plotted in an m dimensional space of vari ables Diagram A shows a plot of the spectra represented by grey points in a three dimensional space of variables as si
35. Peak Variance View context menu offers the following commands Command Used to Coordinates Show Hide the display of cursor coordinates in the status bar Grid Show Hide the grid in the view Scaling Pop up scaling commands for the view Auto Scaling Activate Deactivate auto scaling in the view Zooming Activate Deactivate the zoom in mode in the view Undo Zoom Same as Undo Zoom command from View menu Redo Zoom Same as Redo Zoom command from View menu Display Mode Pop up display modes for the view Background Color Define the background color of the display region of views Show Spectrum Show in the Spectra View the spectrum that corresponds to the right clicked data point ClinProTools User Manual Version 2 2 9 69 Reference Part ClinProTools Context Menus Bruker Daltonik GmbH 9 2 7 X Y Axes Context Menus Right clicking on the x axis or y axis of a view opens a context menu offering the following commands Command Used to Hide Show X Axis Show Hide the x scale of the selected view Hide Show Y Axis Show Hide the y scale of the selected view Axis Font Define the axis font for all views Background Color Define the background color of the axes 9 2 8 Model List View Context Menu The Model List View context menu offers the following commands Command Used to Show Model Show the selected model in the Model report Save Model As Savesthe selected model as XML file with a specified name Remove Model
36. Preparation Reset Settings Data Preparation Figure 9 28 Data Preparation menu Command Used to Settings Spectra Define the spectra preparation and recalibration settings Preparation Settings Peak Define the peak picking and peak calculation settings Calculation Load Settings Data Load the selected data preparation settings XML file Preparation Save Settings Data Save the current data preparation settings in an XML file with Preparation a specified name Reset Settings Data Reset the current data preparation settings to their defaults Preparation Recalibration Recalibrate spectra and calculates average spectra Average Peak List Calculate the average peak list on the total average spectrum Calculation Peak Calculation Calculate peaks and peak statistic in the single spectra 9 1 4 1 Settings Spectra Preparation Command The Settings Spectra Preparation command is used to set the parameters for pre paring spectra by modification and selection and picking recalibration masses during spectra loading as well as for spectra recalibration The settings are stored with the peak calculation settings in the SettingsDataPreparation xml file The command opens the Settings Spectra Preparation dialog Figure 9 29 9 30 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Reference Part ClinProTools Menus Settings Spectra Preparation Resolution 800 Resolution x Baseline Subtraction Top Hat B
37. Sensitivity and Specificity which describe the quality of the model Two types of validation are used by the software Cross valida tion and External validation A 10 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Appendix A 3 Abbreviations ANOVA analysis of variance AUC cc CV FN FNF FP FPF GA GB GUID I O k NN KT npc PCA p value QC RAM rc ROC SNN SRM SVM TIC TOF TN TNF TP TPF XML area under the ROC curve correlation coefficient cross validation false negative false negatives fraction false positive false positives fraction Genetic Algorithm Gigabyte globally unique identifier input output k nearest neighbor Kendall s tau b algorithm multiple measurement number of peak combinations principle component analysis probability value QuickClassifier algorithm random access memory recognition capability Receiver Operating Characteristic also Receiver Operating Curve Supervised Neural Network algorithm structural risk minimization Support Vector Machine algorithm total ion count time of flight true negative true negatives fraction true positive true positives fraction extensible markup language ClinProTools User Manual Version 2 2 A 11 Appendix Bruker Daltonik GmbH A 4 Data Exchange Formats ASCII Import We support Ciphergen ASCII format The format will be detected automatically The ASCII files in the spectra collection folder must all hav
38. The single compared peaks are ordered with decreasing absolute correlation value Like in the Correlation Matrix report Section 8 1 1 3 a color code ranging from red cc 1 to blue cc 1 is used to highlight different cc ranges in the list This allows quickly detecting highly correlated peak pairs gt ClinProt Correlation List Peak Index 30 Mass 2757 15 BRUKER Correlation Calculated Over All Classes ClnProTools Version 2 2 build 28 Correlation Algorithm standard ID Mass 18 1881 17 23 2358 99 a J82 4 4970 04 19 1898 28 50 77747 51 8151 52 21 2023 54 57 9299 65 53 8776 22 54 8821 36 Figure 8 5 Correlation List report section 8 6 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Reporting Data 8 1 1 5 Model List Report The Model List report ClinProtModelList xml Figure 8 6 is created and shown using the Model List command from the Reports menu or by clicking _Model List It lists the parameters of all loaded models in a table The following data is displayed for each model Column Description Name Model name Algo Classification algorithm used Validation Results from cross validation overall and for each class and recog nition capability calculation If cross validation could not be calculated due to not enough spectra this is indicated by Insuf under XVal GA Param GA specific parameter settings filled for GA
39. User Manual Version 2 2 Bruker Daltonik GmbH Installing and Licensing ClinProTools 2 INSTALLING AND LICENSING CLINPROTOOLS Bruker Daltonics ClinProTools 2 2 is supported by Windows2000 and WindowsXP English Version For details on the required service packs see the read me file on the installation CD for system requirements Section 2 1 Working with ClinProTools 2 2 requires the MATLAB Component Runtime application be installed on your system Thus installation first checks if this component is present and if not it prompts you to install the application prior to starting ClinProTools installa tion The ClinProTools software and the MATLAB Component Runtime software are installed from the ClinProTools installation CD delivered Initial installation of ClinPro Tools on a computer automatically creates a temporary license valid for 30 days To work with ClinProTools in future you have to enter the ClinProTools license key you received A separate license is needed for usage of the Support Vector Machine algorithm this license is not part of the 30 day test license 2 1 System Requirements CPU Pentium IV processor equivalent e Clock 3 GHz or more for satisfying data handling double processor machine recommended e Hard disk at least 2 GB of free disk space e Main Memory 2GB RAM up to 4 GB are supported e Operating System Windows 2000 or Windows XP English Version with the latest Service Packs e Internet Explorer 7 or 6
40. a ites aan Vesa 6 22 6 2 4 External Validation eaaa ae aeea ai aaiae daa aaa ii 6 24 6 3 Spectra Classification ceccccecccceceeeeeceecneceeeeeeeeecececaeaeeeeeeesesacqaeaeeeeeeeseesenueeeeeeees 6 25 iv ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Contents 6 4 Statistics in ClINPrOTOOIS 0 ec ee E a E 6 26 GAT Statistical FOSS css sccutecetds decdicottctentts Petthdy E 6 26 OELE Talest ha s teat ieee dni tte eet aera ea 6 26 ETZ ANOVA TV eStie E E tied eatin mean de 6 27 6 451 3 Wilcoxon Test rian e aA ate Aiea 6 27 6 4 1 4 Kruskal Wallis TeSt i cccceeeeeeeeeeeeeeeeeeeeeeeeeseeeeeeseeaeeeseeaeees 6 28 6 4 1 5 Anderson Darling Test cc eeeeeeeeeeee eee eeeeeeeeeeeeeeteeeaeeeeeeeaaees 6 28 TETE PNIU o fans ieee eee dele 6 30 042 Statistical MethOOS ivcciecec seses coectetedstectdinn teeta neeettid seecttinbaeeetea sic ineeeede be 6 30 6 4 2 1 Correlation Analysis cccccceeeeeeeeeeeeeeeeeeeeeeeeseeeeeeseeaeeeeeeaees 6 30 6 4 2 2 Receiver Operating Characteristic eceeeeeeeeeeeeeeteeeteeneeees 6 32 6 4 2 3 Principal Component Analysis cccceesecceceeeeeeseseeeteeeeeeeees 6 34 6 4 2 4 Unsupervised Clustering 0 00 eeeeeeeeeeeeeeeneeeeeteeeeeeeeeeneeeeeeeanees 6 36 6 4 2 5 Pattern Matching for Outlier Detection ee eeeeeeeteeeeneeees 6 36 6 4 3 Remarks on Statistical Problems with MS Data cc eeeeeeeeeeeeeeneees 6 37 6 4
41. and target folder and click Save 3 If you have selected an existing file name answer the confirmation request to overwrite the file To load the model generation settings 1 From the Model Generation menu select Load Settings Model Generation This opens the Load Settings Model Generation File dialog with the SettingsModel Generation folder being opened by default 2 Navigate to the file you want to load Double click it or select it and click Open This overwrites the current data preparation settings with the loaded ones To reset the current model generation settings to the defaults 1 From the Model Generation menu select Reset Settings Model Generation 2 Confirm the appearing request to reset the current settings to the defaults 7 2 1 2 Checking and Optionally Changing the Current Peak Selection Before running the model generation workflow you should check the current peak selection that was set up within the peak calculation workflow and change it if desired All peaks with a blue integration region will be included in model generation whereas peaks with a gray integration region will be excluded Note The peak selection settings may strongly influence the quality of the chosen Classification algorithm In many cases a reasonable reduction of peaks improves the classification performed by the algorithms You can change the current selection by modifying the parameters in the Settings Peak Selection dialog Section 9
42. are higher weighted than dimensions which do not contribute to class separation This procedure of optimizing prototype positions with a combined feature selection is applied iteratively for a predefined upper limit of steps The algorithm may stop earlier if some convergence criteria are reached Parameterization The parameters for the SNN are defined in the Settings Supervised Neural Network dialog Section 9 1 5 2 3 The SNN automatically uses the automatic detection mode to determine the best number of peaks to be integrated in the model Section 6 2 1 5 The Advanced parameters define the prototype determination The Upper Limit of Cycles can be set This number should be chosen with respect to the complexity of the data and can be evaluated considering the views the number of picked peaks and the statistics For complex data sets the SNN may need longer runtime to get good results Typically the value can be taken with defaults The user setting k 1 99 is multiplied internally by 100 hence the number of cycles processing the whole model generation data for one time is given as k 100 Typically at least 1000 cycles should be calculated by the algorithm The Number of Prototypes should be chosen with respect to the number of expected sub clusters in the data and the overall data comple xity The default is suitable in general but an increase of prototypes may sometimes improve the model performance but may also lead to over fitti
43. at a time For this select the folder with the spectra you want to load as a class ClinProTools loads all spectra in a folder and its subfolders as one class and prepares them 2 Start peak statistic calculation using the Peak Statistic command from the Reports menu or ae This runs the spectra recalibration spectra averaging and peak calculation processes on the loaded spectra 3 View the Peak Statistic report Section 8 1 1 2 which opens automatically and ClinProTools User Manual Version 2 2 4 3 Getting Started with ClinProTools Bruker Daltonik GmbH lists all picked peaks with corresponding statistical data ordered according to the sort mode for peak selection Figure 4 1 The Spectra View Section 5 1 1 shows all picked peaks marked by highlighting their integration regions and the 2D Peak Distribution View Section 5 1 3 1 displays the distribution of the two first best separating peaks of the peak statistic Figure 4 1 M ClinProTools Class 1 Normal Class 2 Spiked oly 2h She UL E Sa BH SS foc 138 JE EE HE ad 8 Load Model Name Algo ClinProt Peak Statistic Eile Edit View Favorites Tools Help E gt x aA O Search She Favorites Br ddress E C BDAL ClinProTools_2_2 Files ClinProtStatisticO019 xml ClinProt Peak Statistic Pk 19 1348 Da ClinProTools Version 2 2 build 38 BRUKER Number of peaks 91 J Sort Mode p value tta S Index Mass DAve PT
44. because the classifi cation model is solely based on these prototypes nevertheless it should have good generalization abilities for unknown data in an external validation To fit this needs the SNN allows for metric adaptation which is useful in the search for biomarker candi dates integrates neighborhood cooperation which typically leads to a better generali zation in external validations and is a margin optimizer which similar to the Support Vector Machine is well founded on mathematical theory The SNN based on the ideas from Kohonen s Learning Vector Quantizers is a modi fied version of the Supervised Relevance Neural Gas algorithm It was developed by 6 16 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Basics Barbara Hammer Marc Strickert and Thomas Villmann B Hammer M Strickert and T Villmann Supervised Neural Gas with General Similarity Measure Neural Process ing Letters 21 1 21 44 2005 and is based on the principle of margin optimization As a simple example a checkerboard data set that consists of two classes blue and green with multiple clusters will be considered These two classes exist in a two dimensional data space which is originated by the strange case that the spectra have only two peaks In Figure 6 3 the first dimension x may be created by peak areas from the first peak and the second dimension y by peak areas from the second peak Each point in the figure represents a spectrum
45. cccccccccccceeeeseeeeeceeeaeeaeceeeeeeeseaaeeeseeeeeeees 4 6 4 9 Closing ClIAPTOTOGIS st icsetcet cctect is eto acts ebictea de ete bbb ceae td seat epettetl et ieeiedd sitatinate 4 8 5 CLINPROTOOLS USER INTERFACE cccccccssscccssssecesesceseennesseaesseaeeesaneseeas 5 1 ora Clin PrOTOOIS Win dOWetecAstesaintsaeden e sete rs tice ca tae oes tah eaa tee a enced tatadia tee 5 1 5b Spectra VICW eeecc cerns e seen dieses da terna ace a e hide baie de etn genaee ead 5 2 512 GOUStACK VIEW erecta savavensa dees n Sook cwstota e atresia aaema dinar eats 5 3 51220 GOlVIOW mi secsevsat Heseheateristastestaleas ebstiatasoiartedhe an hated eanitiaceeenedlens 5 4 51 22 Stack VieW eeraa a a aa aaiae 5 5 5 1 3 Peak Statistics VieW niii a a a R A 5 6 5 1 3 1 2D Peak Distribution VieW ccccccceescecceeeceeeeseeeeeeeeeeanenseeeeeeees 5 6 513 2 ROG Curve VIEW teienei aana e vache abdeaaetiastiaua ids OA 5 7 5 1 3 3 Single Peak Variance View cceeccceeeeeteeeeeeeeteeeeeeneeeeeeneeeeeeaas 5 8 5A MOGSIEISCMIGW s ccses aii desth face foit acest ia toihs feces a eas ia aa aa a A 5 9 5 1 5 TROOIDANS EAEE AEE S E E TEE Vien adc Tad GOs GA 5 10 ClinProTools User Manual Version 2 2 iii Contents Bruker Daltonik GmbH 5 6 Status Balin ni cesses te eer Res Oe eine eink 5 10 5 1 7 Altering the ClinProTools Data Plotting Views 0 00 0 ec ccceeeeeeeeeetteeeeeeneeees 5 11 5 1 7 1 Customizing the Display ececc
46. check which model s you want to keep and save it them Section 9 1 1 3 9 50 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Reference Part ClinProTools Menus Shortcut Button _ Clear Al 9 1 5 7 Settings Cross Validation Command The Settings Cross Validation command is used to set the parameters for cross vali dation Cross validation in ClinProTools requires that at least 20 not excluded spectra over all groups are available This also applies to working with groups of spectra from multiple measurements Section 6 1 3 2 here at least 20 groups must be available The settings are stored as described for the GA settings Section 9 1 5 2 1 The command opens the Settings Cross Validation dialog Figure 9 41 Note If you change the cross validation settings when models of the state Calculated are present in the Models List the respective models are reset to the state Added and have to be calculated again This ensures that all models in the list are calculated based on the same cross validation settings Settings Cross Validation IV Calculate Recognition Capability IV Calculate Cross Validation Mode Random C K Fold C Leave One Out Random Parameters 20 Percent to Leave Out 10 Number of Iterations K Fold Parameters K Divide in K Parts Cancel Help Figure 9 41 Settings Cross Validation dialog default setting ClinProTools User Manual Version 2 2 9 51 Referenc
47. classes ClinProTools User Manual Version 2 2 8 3 Reporting Data Bruker Daltonik GmbH Column Description PTTA P value of t test 2 classes Section 6 4 1 1 or ANOVA test gt 2 classes Section 6 4 1 2 range 0 1 0 good 1 bad Preferable for normal distrib uted data PWKW P value of Wilcoxon test 2 classes Section 6 4 1 3 or Kruskal Wallis test gt 2 classes Section 6 4 1 4 range 0 1 0 good 1 bad Preferable for not normal distributed data PAD P value of Anderson Darling test Section 6 4 1 5 gives information about normal distribution range 0 1 0 not normal distributed 1 normal distributed AveN Peak area intensity average of class N StdDevN Standard deviation of the peak area intensity average of class N CVN Coefficient of variation in of class N ClinProt Peak Statistic BRUKER gt ClinProTools Version 2 2 build 28 Number of peaks 71 Sort Mode p value tta S Index Mass DAve PITA PWKW PAD Avel Ave2 StdDevl StdDev2 CV1 CY2 X 19 1347 5 88 73 lt 0 000001 lt 0 000001 lt 0 000001 19 6 108 33 1 44 9 48 Wen ELR X 22 1619 8 49 37 lt 0 000001 0 000001 0 000001 13 15 62 52 1 65 7 53 ORA e K 41 2464 48 40 44 0 000001 0 000001 0 000001 8 28 48 72 0 63 SS 7 56 11 44 X 38 2092 62 77 86 lt 0 000001 lt 0 000001 lt 0 000001 17 2 95 05 1 81 13 03 10 53 13 71 X 16 1296 46 44 33 lt 0 000001 0 000001
48. dialog Section 9 1 5 7 With activated cross validation after each model generation a final cross validation is applied The three modes are illustrated in Figure 6 5 and described in the folowing 6 22 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Basics Whole data set Model generation set Test set here 24 spectra 10 spectra used to generate spectra to be classified diseased hatched pieces the model against the model and 14 non diseased Random Figure 6 5 Illustration of the cross validation modes used in ClinProTools ClinProTools User Manual Version 2 2 6 23 Basics Bruker Daltonik GmbH e Random A random subset of data points taken over all classes is selected and omitted from the model generation procedure The model is calculated with the remaining data points and the random set of data points is classified against the model The obtained classification results are stored This procedure is repeated for a defined number of iterations and finally the averaged classification results give the prediction capability e K Fold The data set is divided into k equal parts of data points Then k models are generated where each time a different one of the k part is omitted The omitted part is classified against the model calculated from the remaining k 1 parts The obtained classification results are stored for the k models averaged and returned as the prediction capab
49. e Model generation validation or classification process the data plotting views remain unchanged You can start the respective process again Shortcut Button or Cance 9 1 1 4 Close All Command The Close All command closes and unloads all spectra and models in order to load new model generation classes Confirm the request to close all spectra Shortcut Button 9 1 1 5 Info Loaded Classes Command The Info Loaded Classes command shows path information about the loaded spectra collections All classes are listed in an automatically opened information box with their corresponding paths numbered with respect to their loading order Figure 9 3 ClinProTools SpectraCollection Paths Class 1 D Data Files ClinProTools ClinProTools Test Data EDTA Run o0h Class 2 D Data Files ClinProTools ClinProTools Test Data EDTA Run O2h Class 3 D Data Files ClinProTools ClinProTools Test Data EDTA Run Oth Class 4 D Data Files ClinProTools ClinProTools Test Data EDTA Run 06h Class 5 D Data Files ClinProTools ClinProTools Test Data EDTA Run O8h Figure 9 3 Path information about loaded spectra classes 9 4 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Reference Part ClinProTools Menus 9 1 1 6 Save Class Paths Command The Save Class Paths command is used to save the paths of the currently loaded model generation classes as ClinProtSpectralmport xml Appendix A 4 This allows loading the referenced spectra v
50. fail but within the remaining multiple measurements of the same sample a sufficiently well measurement exists ClinProTools supports spectra grouping from multiple measurements when the Support Spectra Grouping option is set The user can decide if all valid multiple 6 10 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Basics measurements should be processed or a selection of one characteristic spectrum per sample should be applied In the first case the remaining multiple measurements after some optional quality filter steps are averaged to reduce the overall measurement variance In the second case the similarity filter will be applied It aims on the selection of the most characteristic spectrum within a given set of spectra from the same sample To obtain a useful selection only those multiple measurements should be processed by the similarity filter which can be considered as characteristic and similar for the current sample Therefore a prefiltering using the noise filter and the adduct filter is recom mended Finally the similarity filter returns one spectrum per sample using a mathema tical similarity measure or the spectrum with the highest intensity if only two spectra remained Spectra quality filter Using the spectra quality filter both spectra of low quality and spectra that have bad calibration properties can be detected and excluded As part of the recalibration step a list of masses list of reference
51. import file and click Open This loads all referenced spectra according to their class membership 7 1 3 Manually Excluding Including a Spectrum Spectra can automatically be excluded by applying specific selecting filters Section 6 1 3 2 during spectra loading In addition you can manually exclude spectra you do not want to use in further processing or re include spectra that have automatically been excluded before Spectra can only be excluded or included before any further processing is started Excluded spectra are displayed in the Spectra View and Stack View with a darker color than the included spectra of the same class e g in dark red instead of red In the Gel View excluded spectra can be marked by a default color code Section 9 1 3 7 3 ClinProTools User Manual Version 2 2 7 5 Workflows in Detail Bruker Daltonik GmbH which indicates the reason of exclusion Moreover excluded spectra can be hidden from being displayed in the Gel Stack View Section 9 1 3 7 4 To exclude include a spectrum manually 1 In the Spectra View or Gel View select the spectrum you want to exclude include 2 From the Edit menu select Exclude Spectrum resp Include Spectrum Alternatively you can select the command from the Spectra View context menu 7 1 4 Recalibrating Spectra and Calculating Average Spectra The recalibration workflow performs spectra recalibration if enabled default setting as well as total average spectrum a
52. marks of the plot the so called whiskers indicate the maximum and minimum peak area intensity within a given class The box indicates the 25 quartile bottom and the 75 quartile top and the horizontal intersection denotes the median 50 of the values fall into this interquartile range and the whiskers give you an impression of how much the remaining 50 of the values spread Outliers are not indicated in the standard box plot You can display modified box amp whiskers plots showing outliers by also activating the Spectra View gt Outliers for Box amp Whiskers command The box amp whiskers plots give a graphic representation of homogeneity of the areas of a certain peak in the spectra of one class They allow assessment of the quality of the peaks in a model A peak where the box amp whiskers of the individual classes are well separated with only minimal overlap of the whiskers is better suited for classification than a peak with overlapping box amp whiskers The plot is drawn on a unique scale independent of the peak intensity scale The box amp whiskers plots are hidden by default arb u ools Test Data EDTA RunW0hiSample_E10_1SLin fid 2020 2025 miz Figure 9 143 Standard box amp whiskers plot calculated from the areas intensities of the respective peak in the spectra of five classes Shortcut Button E 9 18 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Reference Part ClinProTools Menus
53. model 7 13 Q Includin 7 9 9 76 REG aie QuickClassifier algorithm 6 12 6 19 Selecting automatically 7 8 QuickClassifier parameters 9 48 Selecting for model generation 7 12 Peak calculation 6 6 7 8 R Peak Calculation command 9 41 Peak calculation parameters 7 2 9 37 Ba c mmand ee Peak calculation workflow 7 8 Recalibration workflow 7 6 Peak distribution Oe ot Receiver Operating Characteristic Peak Distribution command 9 17 Ge 6 32 Bin ae Gated Redo Zoom command 9 13 E ae 8 14 Reference tables A 1 Se ina 7 18 3 3 Remove Model command 9 77 owing eee Remove Peak command 9 77 Peak list export Report CART format 8 14 A 13 AAA T XML format 8 14 A 13 adea S45 Peak List Export command 9 6 Savin g 8 12 Peak list normalization for model AD 8 1 p ae Pi Reports menu 9 60 SaL marker Reset Settings Data Preparation Peak Markers command 9 20 command 9 40 Peak number determination modes 6 20 Reset Settings Model Generation Peak picking 6 4 7 7 command 9 53 On singlespectta oo Reset View Settings command 9 29 On total average spectrum 6 5 Resettin Peak selection parameters 7 2 7 12 9 43 Data Poig views 5 14 Peak statistic calculation 4 3 7 18 Data preparation settings 7 3 Peak statistic calculation standard File open paths 4 2 B panic ae ne General settings 4 2 Se ee Model generation settings 7 11 Peak statistic parameters 9 63 View settings 9 29 Peak Statistic report 7 18 8 3 Resol tion g 6 8 9 32 Peak Statistics View 5 6 ROC curve 5 7
54. of a PC The values for the loadings are between 1 and 1 A negative value indicates a negative loading of the respective variable a positive value reports a positive loading of the variable and a value of 0 shows that the respective variable has not influence on the variability of the ClinProTools User Manual Version 2 2 6 35 Basics Bruker Daltonik GmbH PC In the case of mass spectra the loadings give information about the contribution of single peaks to the variance covered by the respective PC For details please refer to T Jolliffe Principal Component Analysis 2002 Sprin ger 2 edition 6 4 2 4 Unsupervised Clustering A clustering workflow has been added using a hierarchical clustering algorithm The calculation can be done on PCA transformed data or on the untransformed peak lists If the PCA is used limiting the PCs to those necessary for explaining 95 of the variance serves as a good data reduction After performing the calculation the class membership of the data sets is stored as ClinProtClustering xml in the CPT folder With hierarchical clustering in the data space the distance of the data points is calculated based on a metric For a given number of classes the data are grouped accordingly A dendrogram presenting the hierarchy is displayed The complete tree ClinProtClustering tree xml is exported to the CPT folder If the full tree option is set the spectra paths can be displayed at th
55. of broad and overlapping peaks too much For that purpose ClinProTools offers two algorithms each with a parameter to be able to optimize the baseline correction e Convex Hull Baseline This type of baseline algorithm constructs the baseline by fitting multiple parabolas to the spectrum The baseline is then refined in an iterative way This is done in a way that the baseline is at least almost always below the spectrum therefore the name Convex Hull baseline The Baseline Flatness para meter influences baseline construction e Top Hat Baseline This type of baseline algorithm constructs the baseline by means of morphology operators The baseline of the spectrum is obtained in two steps First each data point is replaced by the minimum value of the spectrum within n data points which gives the so called erosion Then within the same number of data points each value is replaced by the local maximum of the minimal values giving the opening of the spectrum which is the baseline The number of data points over which the minimum and maximum value is searched for is a function of the mass The range for the minimum and maximum search can be enlarged with the Mini mal Baseline Width parameter For reference we refer to J Serra Image Analysis and Mathematical Morphology Academic Press New York 1982 The advantage of the Top Hat baseline is the fact that the tuning parameter is giving a more transparent option to modify the bas
56. of spectra or only class paths Appendix A 4 In ClinPro Tools spectra import XML files can be saved via the Save Class Paths command The command opens the Open Spectra Import XML dialog with the ClinProtSpectra Import folder opened by default Navigate to the file you want to load and click Open Loading and preparation of spectra is performed as described with the Open Model Generation Class command After opening a spectra import XML file no additional spectra import XML file can be loaded However you can add further spectra collections via the Open Model Genera tion Class command Shortcuts Button a Keys Ctrl I ClinProTools User Manual Version 2 2 9 3 Reference Part ClinProTools Menus Bruker Daltonik GmbH 9 1 1 3 Cancel Command The Cancel command cancels any currently running spectra loading recalibration peak calculation model generation or classification process The effect of canceling depends on the running process When canceling a e Spectra loading process the model generation class currently being loaded as well as all previously loaded classes are unloaded You have to start loading classes again e Data preparation process spectra recalibration peak calculation the data plotting views become temporarily cleared You have to run the canceled process again to redraw the data in the views Alternatively if you do not want continue processing you can select the Close All command from the File menu
57. on the Total Average Spectrum 6 5 6 1 1 5 2 Peak Picking on the Single Spectra eee 6 5 6 1 1 6 Peak Calculation in the Individual Spectra ccceeeeeeeeeeeees 6 6 6 1 1 7 Normalization of Peak Lists for Model Generation 0 06 6 6 6 12 25 Spectra Group inGseieiscteties beets a e eaa E e a a D3 React acceded nee 6 7 6 1 3 Additional Fite Sss eiri a a aa a aa aaa 6 8 6 1 3 1 Filters Modifying Spectra essseeesseeesrresesrrresrsrniseernnerinnesiennanerenas 6 8 6 1 3 2 Filters Selecting Spectra aassseeessseererreeserraersrneseernnerinnestennaatenane 6 9 6TA Manual Peak Eding reien aae AER A EEEa 6 11 6 2 Model Generation and Validation cccccccceceeeeeeeeceeeeeeeeeeeecaeaeeeeeeesessenencaneeeess 6 12 6 2 1 Classification Algorithms 20 0 0 ce ccceeeeeeeeeeeeeeeeeeeeeeeeeeeaeeeseeeeaeeeseeaeereeeaaeees 6 12 6 2 1 1 Genetic Algorithm oieee ieii ii 6 13 6 2 1 2 Support Vector Machine Algorithm ccceceeeeeseeeeeceeeeeeees 6 15 6 2 1 3 Supervised Neural Network Algorithm eeeeeceeeseeeeeenneees 6 16 6 2 1 4 QuickClassifier Algorithm 0 00 0 cee cceeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeneaeees 6 19 6 2 1 5 Detection Modes to Determine the Best Number of Peaks ina Model aeea aiena ae a e ain eae 6 20 6 2 2 K Nearest Neighbor Classification ccccccceceeeeeeeeeeeseeneeeeeseeneeeeeenanes 6 21 6 2 3 Cross Validation ai ai Aediee ite tice
58. peak Thereby interesting peak means that a peak shows a significant difference between the considered classes in a univariate point of view The Wilcoxon Kruskal Wallis test has less restrictive constraints hence the p value calculation needs a larger set of spectra for a valid p value estimation Therefore if the p value for AD gt 0 05 one should always consider the t test ANOVA instead of Wilcoxon Kruskal Wallis to derive a decision For details please refer to M A Stephens EDF Statistics for Goodness of Fit and Some Comparisons JASA 69 347 pp 730 737 1974 ClinProTools User Manual Version 2 2 6 29 Basics Bruker Daltonik GmbH 6 4 1 6 P Value A p value is the probability that an observed effect is simply due to chance it therefore provides a measure of the strength of an association A p value does not provide any measure of the size of the effect and cannot be used in isolation to inform clinical judgment P values are affected both by the magnitude of the effect and by the size of the study from which they are derived and should therefore be interpreted with caution In particular a large p value does not always indicate that there is no association and similarly a small p value Section 6 4 3 2 does not necessarily signify an important Clinical effect Subdividing p values into significant and non significant is poor statistical practice and should be avoided Exact p valu
59. re a a a aa ce a 9 82 9 3 2 1 Mark Data Points Command ccccccccecccseeeeecsseeeeeeeeeeeeeeneeeeees 9 82 9 3 2 2 ZOOM Commands arruina a a n ana he tee a e 9 83 9 3 2 3 gt Pam COMMANG a a a aa aaa aada a aian 9 83 9 3 2 4 Rotate 3D Command l sasissesseeeeineeernissnrsinrrrrerrrenriiennrinnsnnrrenrreen 9 84 gT POS Me a Ae se ek le a a a a a 9 84 9 3 3 1 Variance COMMANGA irnir a aa ara 9 84 9 3 3 2 Influence Command assseeieeeenneenennsenrsnsrrrenrrrennriienrrnnnrnernnrrenn 9 84 e Peo 7 BN S EN YET a siete E E E AEAEE ATAT 9 85 93AT PGs COMMANG sniedunen 9 85 10 ERROR TREATMENT cccccescceeccesccececececececeseeneceneeeseceeccecccecececeueceuereeseeesees 10 1 Ai APPENDIX O E sctentenettiteedacaabt vcuds A P A E E E R E A 1 A 1 Quick Reference on Menus Commands Tool Buttons and Shortcuts in Siale Aole EA EAEE EE E E E E E A A 1 APRE E EE T E EA EE E EEA E E A E N A 6 ACS Ab re OE E a a a a a oA A 11 AA Data Exchange Formals enea A A EE A EEEE A 12 AO CP arttNUMBONS ern E A cena da aeiad ceca unde e a da decent AE A 14 l NIB AE E E E E N E A A PE E L E E ATTA T E E E A E l 1 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Preface 1 PREFACE The Bruker Daltonics ClinProTools 2 2 application referred to as ClinProTools is an easy to use data post processing software for visualization data reduction data mining and building predictive models from prote
60. region If the selected integration region overlaps with the integration region of an already existing peak adding of the new peak is refused 2780 miz 2780 miz 2780 miz 2780 miz Figure 9 53 Adding a new peak manually using the Distance cursor Mass Range Add a Peak with Integration Limits From 2767 8 to 2791 59 Da Figure 9 54 Mass Range dialog to confirm adding the stated peak 9 2 9 2 Auto Scaling Command The Auto Scaling command activates deactivates the auto scaling mode in the Spectra View or Single Peak Variance View When auto scaling is active the y axis scaling is automatically adjusted to fully display the most intense peak in the current mass range Spectra View resp the maximum statistic value of the current peak in the loaded classes Single Peak Variance View 9 2 9 3 Background Color Command The Background Color command is used to change the background color of the display regions views Selecting this command from the context menu of any of the axes allows changing the background color of the axes of all views The command ClinProTools User Manual Version 2 2 9 71 Reference Part ClinProTools Context Menus Bruker Daltonik GmbH opens the standard Color dialog to select the desired color from a list of Basic colors or defined Custom colors 9 2 9 4 Coordinates Command The Coordinates command shows hides the cursor coordinates in the status bar Section 5 1 6 When the co
61. respect to the base peak in order to be detected Peaks with a lower relative intensity are excluded Single Spectra Check this option if peaks should be picked on the single spectra and averaged peak lists over all classes and the single classes should be calculated Section 6 1 1 5 2 In addition a peak statistic for further use in pattern matching algorithms is stored The averaged peak list over all classes is used instead of the average peak list obtained from the total average spectrum in CPT 2 1 Because overlapping peak ranges from different classes will be cut into separate non overlapping ranges it is preferable to use peak intensities instead of areas which might better represent the different peaks in this case If smoothing in Settings Spectra Preparation dialog is currently not enabled when selecting this option a warning will appear which recommends enabling smoothing You can skip this warning in future by checking the corresponding option in this message and turn it on again by enabling the Show Smoothing Warning option in the the Settings General dialog Signal to Noise Threshold on Single Spectra Enter the minimum signal to noise ratio a peak must have in order to be detected The higher this value the less peaks are detected but the higher is the quality of the detected peaks Reasonable values are 2 0 and above Maximal Peak Number on Single Spectra Enter the maximal number of peaks to pick on a single spectr
62. same power as the distribu tion free test because it does not any longer depends on a specific distribution This has to be kept in mind it may still be true that more relevant peaks have smaller p values than unimportant peaks but the exact p value is not any longer valid 6 4 3 3 Multiple Measurements of the Same Sample Multiple measurements mm occur if the same sample is measured multiple times This can automatically be done by the ClinProt measurement system by multiple spots of the same sample The obtained spectra generally 4 for each sample are stored in a common directory named by the sample_id These measurements should in general be very similar ClinProTools has to handle samples measured with mm in a special manner for e g formal reasons regarding statistics and model building If mm are available for a sample the Support Spectra Grouping option in the Settings Spectra Preparation dialog has to be activated before opening any files This enables ClinProTools to search for specific directory structures for mm as created by the measurement system By default the measurement system automatically manages the directory structure and mm are supported valid Note Therefore it is strongly recommended not to modify the directory structure under a sample set top folder Otherwise invalid groupings may occur 6 40 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Basics If mm are present an additional process
63. system we recommend to make use of the 3GB flag only in combination with full 4 GB RAM For detailed information also about other operation systems please refer to the Microsoft web page at http www microsoft com whdc system platform server PAE PAEmem mspx 2 4 Licensing ClinProTools Installation of ClinProTools on a computer automatically creates a temporary license valid for 30 days To work with ClinProTools in future enter the license key you received If you ordered the Support Vector Machine license also enter the corre sponding license key Permanent as well as temporary licenses can be given In the latter case an expiration warning will inform you about the forthcoming expiration firstly 30 days before the license will expire License keys have to be entered in the Bruker Daltonics LicenseManager which you can launch from the Windows Start menu The ClinProTools User Manual Version 2 2 2 3 Installing and Licensing ClinProTools Bruker Daltonik GmbH LicenseManager will list any license currently available for any Bruker Daltonics application Alternatively you can open the LicenseManager from ClinProTools using the LicenseManager command from the Compass menu Note If the license key for the Support Vector Machine is entered when ClinProTools is started a restart of ClinProTools is necessary to make the Support Vector Machine available To license ClinProTools and Support Vector Machine optionally cio MSE
64. test or within ClinProTools an evaluation of the discrimination quality of a peak The ROC curve is an exploration of what happens to the true positives and the false positives if the position of an arbitrary threshold is varied This arbitrary cut off point splits the values into a fraction representing a positive test result values above the point and a fraction representing a negative test result values below the point In ClinProTools ROC curves can only be generated for the case of two model generation classes because a true false decision is not possible for more than two classes Recognition capability The recognition capability is one measure to describe the performance of a Classifier It is calculated for a determined model as the relative number of correct classified data points by the classifier for the given model under the constraint that all tested data is previously used for the determination of the model or training of the classifier In other words the recognition capability indicates how good a determined model is able to classify the data which is used for model generation If the recognition capability is low the classifier was not able to learn the underlying data characteristic This may happen if it is not possible to determine a relation between the properties of the data and the given labeling A high recognition capability however does not necessarily mean that the data are separable or the model is good If e g al
65. the GA is that it needs much less computational time than the brute force approach while still yielding good results The drawback is that you obtain only a near optimal solution since you cannot guarantee to find the best combination if you do not test all of them How the GA works The GA works on a population which consists of a multitude of peak combinations During selection the fittest peak combinations are chosen and the less capable are abandoned This is done by optimizing a cost function which aims on optimal class separation with variance high between classes Using the cost function each peak combination is rated by an expense factor which is used as a measure for the fitness The crossover combines randomly selected pairs of peak combinations to produce child peak combinations which replace their parent peak combination The intention here is to combine two fairly good peak combinations to form even better ones Of course crossover of peak combinations can also result in less fit combinations but these will not survive for a very long time Finally a small amount of peak combinations is modified randomly during mutation This is done to keep genetic diversity and to prevent a premature convergence to a local optimum The expectation is that the average fitness of all peak combinations rises and the best fitness observed will improve Parameterization The basic and advanced parameters for the GA are defined in the Settings Genet
66. the chosen classification algorithm Settings Peak Calculation Peak Picking on Resolution Used From Settings Spectra Preparation Dialog Total Average Spectrum C Single Spectra Signal to Noise Threshold Signal to Noise Threshold on Average Spectrum on Single Spectra ico Maximal Peak Number Relative Threshold Base Peak on Single Spectra on Average Spectrum Peak Aggregation Limit Peak Number y Minimal Occurrence in Single Spectra Maximal Peak Number ppm Aggregation Width Sort Mode Peak Calculation C Use Areas Use Intensities Integration Type Cancel Help Figure 9 32 Settings Peak Calculation dialog default setting In Peak Picking define how to pick peaks Total Average Spectrum Check this option if peaks should be picked on the total average spectrum and the overall average peak list should be calculated from this spectrum like in ClinProTools 2 1 Section 6 1 1 5 1 ClinProTools User Manual Version 2 2 9 37 Reference Part ClinProTools Menus Bruker Daltonik GmbH Signal to Noise Threshold on Average Spectrum Enter the minimum signal to noise ratio a peak must have in order to be detected The higher this value the less peaks are detected but the higher is the quality of the detected peaks Reasonable values are 2 0 and above Relative Threshold Base Peak on Average Spectrum Enter the minimum relative intensity a peak must have with
67. the peak distribution symbols remains constant when resizing the window The symbols of one class are slightly hori zontally displaced to avoid drawing several symbols in the same place and thus giving a wrong impression of density To display data of another peak you can use the view s scroll bar to browse through the peaks or right click the desired peak it in the Spectra View and use the Variance for Peak n command Section 9 2 9 24 5 8 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH ClinProTools User Interface 5 1 4 Model List View The Model List View Figure 5 8 lists all models currently available in ClinProTools in a tabular format the model list It also offers various buttons to quickly launch model related commands Each model is displayed along with corresponding data Model Name classifier Algorithm Cross Validation and Recognition Capability results current State and Date Time of calculation The data available for a model depends on the model s current State An added but still not calculated model has the state Added The state Calculated indicates an already calculated model from the current session and the state Loaded a loaded formerly saved model The corresponding cross validation and recognition capability results are shown as well as date time of model calculation When a model is currently under calculation the progress of model generation and validation is shown as Generating M
68. the request to remove ClinProTools from your system PIO oS eS ClinProTools User Manual Version 2 2 2 5 Installing and Licensing ClinProTools Bruker Daltonik GmbH 2 6 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Data Acquisition for Clinical Proteomics 3 DATA ACQUISITION FOR CLINICAL PROTEOMICS 3 1 Introduction A direct MS analysis of very complex mixtures such as many biological fluids blood serum blood plasma etc can often end up with unsatisfying spectra quality Highly concentrated components may suppress minor components similar mass to charge m z ratio peptides and proteins may result in overlapping peaks Those features can be avoided when the samples are subjected to prefractionation prior to MS analysis A selective enrichment of specific peptides protein fragments and proteins according to their biological chemical or physical properties can improve spectra quality significantly Bruker offers an off line system for enrichment prefraction ation based on magnetic microbeads with different functionalized surfaces The handling of the magnetic beads is simple They are provided as different kits ClinProt Kits and each kit contains a detailed protocol for sample preparation opti mized on blood serum Additionally the flexible handling of the beads enables the user to vary the protocols optionally and to adapt them to their special tasks scaling concentration variation multi s
69. the selected spectra collection for model Class generation Open Spectra Import Open the selected spectra import XML file and load the XML referenced spectra Cancel Cancel any current loading calculation model generation classification process ClinProTools User Manual Version 2 2 9 1 Reference Part ClinProTools Menus Bruker Daltonik GmbH Command Used to Close All Close and unloads all spectra and models Info Loaded Classes Show path information about the loaded spectra collections Save Class Paths Save the paths of the loaded model generation classes as spectra import XML file Print Print a graphic of the active data plotting view Print Preview Preview the graphic to be printed for the active data plotting view Print Setup Set up the printer and printing options Peak List Export Export the peak list in XML or CART format Browse ClinProTools Browse the ClinProTools folder Folder General Settings Define general ClinProTools settings Exit Close ClinProTools 9 1 1 1 Open Model Generation Classes Command The Open Model Generation Class command is used to open a model generation class ClinProTools loads all spectra in a folder and its subfolders recursively as one model generation class ClinProTools supports loading spectra of the X Mass BAF und ASCII Appendix A 4 file formats For loading ASCII file formats the null spectra exclusion filter Section 6 1 3 2 has to be disabled For model generation y
70. to Copy Copy a bitmap and or a metafile graphic of the selected data plotting view to the clipboard according to the states of the graphic format commands Exclude Include Exclude Include the selected spectrum manually Spectrum Bitmap to Clipboard Activate Deactivate the bitmap format for copying graphics to the clipboard Metafile to Clipboard Activate Deactivate the metafile format for copying graphics to the clipboard 9 1 2 1 Copy Command The Copy command copies a graphic of the selected data plotting view to the clip board This allows pasting that graphic into another application By default ClinPro Tools copies graphics as a bitmap with a resolution of 800 600 pixels Alternatively ClinProTools can set up a metafile with a resolution of 8000 6000 pixels Whether a bitmap and or a metafile are is created depends on the settings of the Bitmap to Clipboard and Metafile to Clipboard commands from the Edit menu If both types are activated the program that pastes the clipboard s contents into its document deter mines which of these formats it uses ClinProTools User Manual Version 2 2 9 9 Reference Part ClinProTools Menus Bruker Daltonik GmbH Shortcuts Button a Keys Ctrl C 9 1 2 2 Exclude Include Spectrum Command The Exclude Spectrum command excludes the selected spectrum from further processing The Include Spectrum command includes the selected manually or auto matically spectrum The command avail
71. to run for optimizing the prototype positions This number should be chosen with respect to the complexity of the data and can be evaluated considering the views the number of picked peaks and the statistics Automatic Detection of Prototype Number Check this option if automatic detection of prototype number should be applied Uncheck it if a fixed number of prototypes should be used and specify the number in Number of Prototypes Number of Prototypes Enter the number of prototypes to use if automatic detection should not be applied The number of prototypes should be chosen with respect to the number of expected sub clusters in the data set and the overall data complexity 9 1 5 2 4 Settings QuickClassifier Dialog The Settings QuickClassifier dialog Figure 9 39 defines the sort weight mode for the QC The QC automatically uses automatic peak detection Section 6 2 1 5 to determine the best number of peaks to be integrated in the model maximum is 25 peaks This setting is stored as described for the GA settings Section 9 1 5 2 1 9 48 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Reference Part ClinProTools Menus Settings QuickClassifier The number of peaks in the model will be detected automatically 1 25 peaks Sort Mode C Difference Average P Value T Test ANOVA C P Value Wilcoxon Kruskal Wallis Figure 9 39 Settings QuickClassifier dialog default setting Sort Mode Select the sort mode us
72. to use Internet Explorer 6 0 or Higher Uses Internet Explorer Requires version 6 0 or higher installed Firefox 1 5 or Higher Uses Firefox Requires version 1 5 or higher installed ClinProTools User Manual Version 2 2 9 7 Reference Part ClinProTools Menus Bruker Daltonik GmbH Suggest Model Name as File Name Check this option if the name entered during adding a new model to the model list should be suggested as ModelName xml in the Save Model As dialog Section 9 2 9 19 Force Entering Model Name Check this option if a new model should be added to the model list only if a model name has been entered Check Memory on Load Check this option if it should be checked on spectra loading whether the available memory is sufficient to load the selected spectra If the needed memory size exceeds the available memory the machine might slow down or come to a standstill The memory size is rated as insufficient when needed memory x 2 gt available memory In this case a warning message is launched which asks you whether to continue You can set an option to skip this message in future Check Memory for PCA Check this option if it should be checked on PCA start whether the available memory is sufficient to perform PCA on the loaded data set s If the needed memory size exceeds the available memory a warning message is launched Classify in Batch Mode Check this option to activate the batch mode and uncheck it to act
73. unsupervised clustering the separation into classes will be ignored i e all data will be treated as one group The command opens the Unsupervised Clustering dialog Figure 9 46 to define the settings for unsupervised clustering and start clustering Unsupervised Clustering V Normalize Data JV Use PCA Data V Reduce Dimensions 95 0 Sum Explained Yariance Create Full Tree V Max Path Length 6 Number of Classes Show ClinProtClustering xml i Cancel Help Advanced lt lt Euclidean Distance Method Minkowski Exponent Average Linkage Method Figure 9 46 Unsupervised Clustering dialog default setting 9 58 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Reference Part ClinProTools Menus Normalize Data Check this option if the data should be normalized before running an unsupervised clustering Use PCA Data Check this option if PCA transformed data should be used Reduce Dimensions Check this option if only the first PCs necessary to explain a part of the variance should be regarded The respective part of variance has to be specified under Sum Explained Variance Sum Explained Variance Enter the minimum percentage of the sum of the variances the first PCs must have Create Full Tree Check this option if the hierarchical clustering should create the complete dendrogram instead of a given maximum number of classes Show Paths For Create Full Tree checked check this opt
74. used to close the current classification This removes the current classification result from the memory In standard mode it also unloads the classified spectra and removes them from the ClinProTools GUI Shortcut Button af 9 56 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Reference Part ClinProTools Menus 9 1 7 Statistical Analysis Menu The Statistical Analysis menu offers the following commands Figure 9 44 Statistical Analysis PCA Unsupervised Clustering Figure 9 44 Statistical Analysis menu Command Used to PCA Perform a PCA on the loaded spectra Unsupervised Clustering Perform an unsupervised clustering on the loaded spec tra 9 1 7 1 PCA Command The PCA command is used to run the PCA workflow performing a PCA Section 6 4 2 3 on the non excluded spectra of the loaded spectra data set s This calculates a PCA and shows the PCA results in the PCA window All generated PCA data is stored as ClinProtPCA xml file in the ClinProTools folder A PCA requires two valid spectra with three peaks being available at least The command can be applied to several classes or to only a single class However in the context of PCA the separation into classes will be ignored i e all data will be treated as one group The command opens the PCA dialog Figure 9 45 to define whether the data should be normalized before running the PCA V Normalize Data i Cancel Help Figure 9 45 PCA dialog def
75. value is a probability measure for the strength of an association between the different classes The exact value and the reliability of the p value depend on some aspects which are explained in the following The statistical tests differ in their performance and their requirements with respect to the underlying data in our case the peak areas As a result not each test can from a theoretical formal point of view be applied to each set of data In addition the per formance power of the test depends on a specific combination of constraints In general each of the supported tests makes at least the following constraints e C1 number of classes 2 classes 2 2 classes e C2 kind of distribution of the underlying data normal distribution arbitrary distribu tion but the same for each class e C3 number of necessary disjunctive samples small large e C4 number of features small large e C5 properties of the measurement To constraint C1 The first constraint is a strict constraint in the sense that some tests are not applicable if the number of classes e g control cancer1 cancer2 is larger than 2 This is the 6 38 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Basics case for the t test and the Wilcoxon test which are only applicable for two class scenarios In general one can say that as larger the number of classes as more com plicated the test scenario To constraint C2 If the da
76. 0 9 1 5 4 Cancel COMMANG ccccceceeeeeeteeeceeeeeseeeceeeeeeeeeesetsssinaeeeeees 9 50 9 1 5 5 Load Model Command ccececeececeeeeeeeeeeeeeaeeeeeeesetsesnaeeeeeees 9 50 9 1 5 6 Clear All Command ccccceceeeccce cee aaa aerea 9 50 9 1 5 7 Settings Cross Validation Command 0 00 0 eeeeeeeeeeteeeeetteeeteeneeees 9 51 9 1 5 8 Load Settings Model Generation Command eeeeeeeeeees 9 53 9 1 5 9 Save Settings Model Generation Command ceeeceeeeees 9 53 9 1 5 10 Reset Settings Model Generation Command l 9 53 9 1 6 Classification Men nenene aotr area ata aae RAEE aE ar aE 9 54 9 1 6 1 Classify Command sranna a a at 9 54 9 1 6 2 External Validation Commangd cccccccecceceeeeeeeseeeesteeeeeeees 9 55 9 1 6 3 Save Classification Command ccccceeeececeeeeeeeeeeteetteeeeeeees 9 56 9 1 6 4 Show Classification Command ccccccecceeeeeeeeseeeeeteeeeeeees 9 56 9 1 6 5 Close Classification Command ccccceeecceeceeeeeeeteetteaeeeeees 9 56 9 1 7 Statistical Analysis MON eecceeeeeceeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeseeaeeeeeneaaees 9 57 9 14 71 PCA Commandin eae det eas 9 57 9 1 7 2 Unsupervised Clustering Command eeeeeeeeeteeeeeeenteeeeeenaeees 9 58 98 JREPOFS Men s 2 454 ecane ea aa oa ie ext tonedtaa eis ote ed aav Ar aE 9 60 9 1 8 1 Spectra List COMMANG ce eeeee cece ee teeeeeeeeaeeeeeeeteeeneaeaeeeeees 9 61 9
77. 1 5 1 For example if you do not want to include all peaks in model generation default setting you can restrict the peaks to be taken this selects only the best peaks according to the chosen sort mode Additionally or alternatively you can change the current selection by manually excluding including peaks Section 7 1 7 Moreover you can force a certain peak s into the model to be generated Section 7 2 1 3 7 12 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Workflows in Detail 7 2 1 3 Forcing a Peak into a Model After running the peak calculation workflow you can force a peak into a model which means the respective peak must be incorporated in the generated model A forced peak is marked by a green integration region before as well as after model generation Forcing a peak into a model can be canceled by selecting the command for the respective peak again To force a peak into a model 1 In the Spectra View right click in the integration region of the peak you want to force and select Force Peak n into Model 7 2 1 4 Calculating a Model After entering a model parameter set in the model list you can calculate a correspond ing model using the Calculate command from the Model Generation menu This command runs model calculation automatically on all models of the state Added currently present in the model list If the loaded spectra are not fully prepared when launching model calculation first the req
78. 1 8 2 Peak Statistic Command 0ccccecceeeeeeeeeeeeeceeeeeeeeseteesteaeeeenes 9 61 9 1 8 3 Correlation Matrix Command ccceceeeeeceeeceeeeeeeeteteeetaeeeeees 9 61 9 1 8 4 Model List Command cccccecceccececeeeeeeeeccecaeeeeeeesesensuaeeeeeees 9 63 9 1 8 5 Settings Statistic Command ce eee eeeeeeeeeeeeeeeeeenaeeetenaeees 9 63 9 19 Compass Men seere errea a ae rae a ETa tre aaa tants sand eaea eir agadan 9 65 9 1 9 1 LicenseManager Command cceeececeeeeeeeeeeeneeeeseeneeeeennaeeees 9 65 9 T10 Help Men mirin reetan aaar a aes aea E ote Att A e 9 65 9 1 10 1 Help Topics COMMANG cceceeeecee cece eeeeeeeeeecaeeeeeeesesensieaeeeeees 9 66 9 1 10 2 About ClinProTools COmmMand ccccceeeeeeceeeeeeeeteeesetaeeeeeees 9 66 9 2 ClinProTools Context MenUuS ccceeeeseeceeeeeeeeeceneaeceeeeeeeseesaaeaeeeeeeeseesnnasaeeeeeess 9 67 9 2 1 Spectra View Context MENnu cceecceceseeeceeeeeeeeeeeseeeeeeeseeaeeeteenaaeeeeeeanees 9 67 9 2 2 Gel View Context MONU a ae a aeaa aa Eaa aa aE 9 68 9 2 3 Stack View Context MenU sisina eaa aaa aa aaa 9 68 9 2 4 2D Peak Distribution View Context Menu c ccseceeeeeeeeeeeeeeeeeeeeeeeetees 9 68 9 2 5 ROC Curve View Context Menu cc cccceecceceeeceeeseeeeeeeeseeneeeeteenaeeeteeaees 9 69 9 2 6 Single Peak Variance View Context MeNu ccc ceeeeeeeeeneeeeteettteeeeeeeeees 9 69 9 2 7 XN Axes Context M
79. 17 Remove Peak N Command The Remove Peak n command removes the selected peak from the average peak list Removing peaks is possible after average peak list calculation as well as after peak statistic calculation or model generation In the latter two cases however the current peak calculation will be reset which is indicated in that the integration regions of all peaks change to gray color This requires running the peak calculation resp model generation workflow again 9 2 9 18 ROC Curve for Peak N Command The ROC Curve for Peak n command displays the ROC curve Section 6 4 2 2 for the selected peak in the ROC Curve View The command is only enabled if this view is active Whether class 1 or class 2 is currently treated as positive depends on the decision made when switching to ROC Curve View via the Peak Statistics View gt ROC Curve command from the View menu ClinProTools User Manual Version 2 2 9 77 Reference Part ClinProTools Context Menus Bruker Daltonik GmbH 9 2 9 19 Save Model As Command The Save Model As command is used to save the selected model in an XML file with a specified name The command opens the Save Model dialog with the ClinProtModels folder as the default storage location Enter a new or select an existing model name and click Save If you have selected an existing name answer the confirmation request to overwrite the file Shortcut Button _ Save 9 2 9 20 Scaling Command The Scaling po
80. 2 2 Bruker Daltonik GmbH Workflows in Detail A list of all loaded spectra can be viewed in the Spectra List report Section 8 1 1 1 using the Spectra List command from the Report menu The report also informs about the spectra s current include exclude state and certain data acquisition parameters 7 1 2 1 Opening a Model Generation Class To open a model generation class you have to select the folder which contains the spectra you want to load as one class The loading procedure has to be repeated for each model generation class of interest To open a model generation class 1 From the File menu select Open Model Generation Class or click 5 2 In the Browse for Folder dialog navigate to the folder that contains the spectra you want to load and click OK This loads all spectra in this folder and perhaps subfolders as one model generation class 3 Ifa message about a too low memory size appears decide how to continue 4 Repeat steps 1 to 3 for each model generation class you want to load 7 1 2 2 Opening a Spectra Import XML File Opening a file of the ClinProtSpectralmport xml format Appendix A 4 allows loading a list of referenced spectra of different model generation classes at once The import file can contain a path list of spectra or only class paths To open a spectra import XML file 1 From the File menu select Open Spectra Import XML or click ar 2 In the Open Spectra Import XML dialog select the desired XML
81. 3 1 Common Statistical Pitfalls Generic Remarks eee 6 37 6 4 3 2 Small P Value Phenomenon cceeeeeteeeeeeeeeeeeeeeeteeteeeeeeaeees 6 38 6 4 3 3 Multiple Measurements of the Same Sample 0ceee 6 40 6 4 3 4 Dependent Measurements of Different Samples from the Same Clinical Perso M anien e e AE 6 41 6 4 3 5 Multiple Hypothesis Testing Analyzing a Large Number of Peaks at the Same Time c cc cceececceeeeeneeeeeeeneeeeeeeneeeeeeneeeeseaaes 6 41 6 4 3 6 How to Determine Sensitivity and Specificity from External Validation z ieaie n tein ea ee 6 42 T WORKFLOWS IN DETAIL iiss asia sccinss issssseaiescaenesssvsseeus sdves suces shee inoata isapan iaar vasen sedans 7 1 7 1 Spectra Loading and Data Preparation ccccccecceceeeeeeeeeeeeeeceeeeeeesecetaeeeeeeeeeeeees 7 1 7 1 1 Defining the Data Preparation Settings ecceeeeeeeeieeeeeeetieeeeeenieeeereee 7 1 7 1 1 1 Setting the Spectra Preparation Parameters ccceeerees 7 1 7 1 1 2 Setting the Peak Calculation Parameters c cceceeeeeteeereeeee 7 2 7 1 1 3 Setting the Peak Selection Parameters cccecseeteeeeeseeeeeee 7 2 7 1 1 4 Saving Loading and Resetting the Data Preparation Settings 7 3 7 1 2 Loading Spectra in CliNProTOOIS e eee eseeeeeeeeeeeeeteneeeeeetteeeeetaeeeerene 7 4 7 1 2 1 Opening a Model Generation Class 0 ccceeeeeeeeeeeteeeeeeeteeeeeeee 7 5 7 1 2 2 Opening a Spe
82. ASICS ON DATA PREPARATION MODEL GENERATION AND SPECTRA CLASSIFI CATION WITH CLINPROTOOLS The following sections provide basic information on data preparation model generation and validation and spectra classification in ClinProTools In addition the used statistical tests and methods as well as certain statistical problems with MS data are described 6 1 Data Preparation ClinProTools uses a standard data preparation workflow including spectra pretreat ment peak picking and peak calculation operations ClinProTools automatically picks the peaks either on the calculated total average spectrum or alternatively on the single spectra and sets up the corresponding peak list After the automatic picking it is possi ble to edit the peaks manually The result of data preparation is a collection of peak areas resp maximal intensities for each spectrum For all spectra the areas maximal intensities of the same peaks are calculated so that for each spectrum the same number of peak areas intensities is obtained ClinProTools supports grouping of spectra from multiple measurements Furthermore the standard workflow can be supplemented by applying additional filters to modify spectra reduce data and exclude spectra of lower quality from further processing For nearly all data preparation steps there are parameters which can be chosen to adapt to the kind of spectra used and to control the number of peaks taken into account The data preparation p
83. C Boldrick Multiple Hypothesis Testing in Microarray Experiments Statistical Science Vol 18 1 pp 71 103 2003 6 4 3 6 How to Determine Sensitivity and Specificity from External Validation The sensitivity of a binary two class classification algorithm such as a blood test to determine whether a person has a certain disease is a parameter that expresses something about the test s performance The same applies to the specificity The semantic of these two characteristics depends on the setting of the positive and the negative class In a binary scenario the corresponding values can be derived from the ClinProTools result output as follows Workflow gt Create a binary classification model gt Click Validate or select the External Validation command from the Classification menu or the Model List View context menu gt Load external data for both classes and start external validation gt Three new XML files are shown Validation Classification Validation class 1 and Classification Validation class 2 reports gt Read from the confusion matrix in the Validation report Section 8 1 1 7 the number of true positives false negatives true negatives and false positives and calculate sensitivity specificity see the following example Determination of sensitivity specificity and positive negative prediction values If we assume the positive class e g diseased is class 1 and the negative class e g control is cl
84. If the Resolution parameter is chosen too large more and more artificial peaks spikes will be found On the other hand smaller Resolution values will remove more and more unresolved shoulder peaks from the peak list Mass range filter To limit the mass range of the spectra to be analyzed you can specify a minimum and maximum mass Otherwise define a mass range that is larger than the experimental mass range which should be the case if you keep the default values 6 8 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Basics Smoothing filter For data smoothing the Savitsky Golay algorithm is available The idea of this algo rithm is to calculate polynomials in the neighborhood of each data point to get a smoothing of the data This can be formulated as M Ji Yo ceyire k A I with coefficients ck The parameter M in the formula is calculated from the given m z smoothing width which can be changed within the Settings Spectra Preparation dialog The number of smoothing cycles can also be chosen which gives the option to apply this smoothing filter multiple times The weights for Savitsky Golay are obtained by considering a least square problem for 2M 1 nodes For details please refer to M U A Bromba and H Ziegler Analytical Chemistry 53 pp 1583 1586 1981 and to A Savitzky and M J E Golay Analytical Chemistry 36 pp 1627 1639 1964 where also tables of the c s can be found Data r
85. Machine dialog Section 9 1 5 2 2 e SNN in the Settings Supervise Neural Network dialog Section 9 1 5 2 3 e QC in the Settings QuickClassifier dialog Section 9 1 5 2 4 Then click OK 4 In the Model Name dialog Section 9 1 5 2 5 enter the name for the new model if desired Click OK to enter the new model parameters set with the specified name in the model list getting the state Added If the Force Entering Model Name option is active and you have not entered a model name a message informs you that a model name is needed Quit the message enter a name and click OK 7 2 1 1 2 Setting the Cross Validation Parameters ClinProTools supports three kinds of cross validation Section 6 2 3 that can be chosen in the Settings Cross Validation dialog Section 9 1 5 7 It is strongly recom mended to apply one kind of cross validation to verify that the obtained models give valid results on unseen data With activated cross validation after each model genera tion a final cross validation is applied You must keep in mind that cross validation in ClinProTools requires at least 20 non excluded spectra over all classes being avail able This also applies to working with groups of spectra from multiple measurements here at least 20 groups must be available You can use the default parameters or specify own settings suitable for your data Alternatively you can load a model generation settings file or reset the current settings to the default va
86. OC Curve and Single Peak Variance Views in various ways to adapt the views to your needs 5 1 7 1 Customizing the Display The display of a data plotting view can be customized using the commands from the context menu of the view s display region or axes A changed setting applies to the selected view only e To show hide the cursor coordinates for a view in the status bar activate deactivate the Coordinates command This does not apply to the Stack View e To show hide the grid in a view activate deactivate the Grid command This does not apply to the Stack View e To change the background color of the display region of a view select the Back ground Color command and choose a new color This does not apply to the Gel View e To change the background color of axes of a view select the Background Color command and choose a new color e To change the axis font of a view select the Axis Font command and choose a new font e To show hide the scale of the x or y axis of a view select the Show Hide X Axis or Show Hide Y Axis command respectively ClinProTools User Manual Version 2 2 5 11 ClinProTools User Interface Bruker Daltonik GmbH 5 1 7 2 Changing the Display Range The display range of a certain data plotting view can be as follows Slave master behavior of x axes of Spectra View and Gel Stack View The x axis of the Spectra View and the x axis of the Gel Stack View show a slave master behavior by default When t
87. OVA Sorts the peaks by the p value from t test Section 6 4 1 1 ANOVA test Section 6 4 1 2 P Value Wilcoxon Kruskal Wallis Sorts the peaks by the p value from Wilcoxon test Section 6 4 1 3 Kruskal Wallis test Section 6 4 1 4 Peak Statistic Creates and shows the Peak Statistic report Section 8 1 1 2 same function as the Peak Statistic command from Reports menu OK Changes the current statistic settings If there are changes that concern the number of peaks for which statistical data is shown the Spectra View is updated accordingly 9 64 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Reference Part ClinProTools Menus 9 1 9 Compass Menu The Compass menu offers the following command Figure 9 50 LicenseManager Figure 9 50 Compass menu Command Used to LicenseManager Launch the Bruker Daltonics LicenseManager 9 1 9 1 LicenseManager Command The LicenseManager command is used to view add and delete licenses for Bruker Daltonics applications It opens the Bruker Daltonics LicenseManager dialog Figure 2 1 showing all licenses currently present for Bruker Daltonics applications To add a new license enter the license key you received in New license key and click Add A new line is added to Existing licenses with the key you have entered the product name and the date until the license will be valid To delete an existing license select it in Existing licenses click Delete and confirm t
88. Remove the selected model from the view Edit Model Name Edit the model name for a parameter set of an added model parameter list Classify Same as Classify command from Classification menu External Validation Same as External Validation command from Classification menu Show Error Show the Error report for a model s ERROR entry 9 2 9 Commands Available from Context Menus Only The following section describes commands available from context menus only in alphabetical order 9 2 9 1 Add Peak Command The Add Peak command is used to manually add a new peak to the average peak list Adding peaks is possible after average peak list calculation as well as after peak statistic calculation or model generation In the latter cases however the current peak calculation will be reset which is indicated in that the integration regions of all peaks change to gray color This requires running the peak calculation resp model generation workflow again 9 70 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Reference Part ClinProTools Context Menus The command displays the distance cursor Figure 9 53 Move the vertical cursor lines as described with the Distance command Section 9 2 9 10 to the positions where the peak should start and end and click the right mouse button Confirm the appearing request to add a peak with the stated integration region Figure 9 54 This displays the peak with a gray colored integration
89. Reports ClinProTools offers various types of reports for showing specific data Section 8 1 1 All ClinProTools reports are created as XML files except the Error report which is of the txt format They can be opened with either the Microsoft Internet Explorer or Excel the respective application can be chosen in the General Settings dialog Section 9 1 1 12 Multiple reports can be open at a time All XML files contain style sheet references which transform them into HTML when opened with a web browser Microsoft Internet Explorer 6 0 is strongly recommended The referenced style sheet must be in the same folder as the XML file To ensure that Excel parses the XML files with style sheet properly make sure that a dot is used as decimal separator in Excel To enforce this go to the Tools Option dialog in Excel On the International tab at Number handling uncheck Use system separators enter a dot as Decimal separator and a comma as Thousands separator If this is not set numbers may be parsed as dates and the like The XML files are stored with a consecutively numbered default name e g ClinProt Statistic0001 xml ClinProtValidation0001 xml in the ClinProTools folder The corre sponding style sheets suffix xsl have been installed there by the setup These XML files will stay in this folder as long as you do not delete them either by automatically removing all temporary XML files Section 4 3 or manually removi
90. TA PWKW PAD Avel Ave2 StdDevl StdDev2 CV1 CY2 X 19 1347 8 88 15 lt 0 000001 lt 0 000001 lt 0 000001 15 38 103 53 1 06 3152 6 89 9 19 X 23 1620 12 47 16 lt 0 000001 lt 0 000001 lt 0 000001 14 14 61 3 1 89 6 55 13 39 10 68 X 42 2465 07 40 39 lt 0 000001 0 000001 lt 0 000001 7 06 47 45 0 61 5 41 8 61 11 39 X 16 1296 73 42 26 lt 0 000001 lt 0 000001 lt 0 000001 6 39 48 65 0 53 6 59 8 32 13 56 Figure 4 1 4 4 2 Basic Workflow Model Generation The basic workflow Model Generation can be used to quickly calculate models using ClinProTools defaults settings This workflow includes spectra recalibration and aver age spectra calculation peak calculation and model generation based on the selected classification algorithm Data of all models present in the model list or of a selected model can be shown in the Model List report or the Model report using the corre sponding command This stores the respective data as ClinProtModelList number xml file or ClinProtModel number xmI file respectively 4 4 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Getting Started with ClinProTools To run the Model Generation workflow 1 Load the two classes Normal and Spiked from the ClinProTools Test Data folder on the installation CD using the Open Model Generation Class command from the File menu or Z One class can be loaded at a time
91. View and disabling it resets to the default behavior ClinProTools User Manual Version 2 2 9 23 Reference Part ClinProTools Menus Bruker Daltonik GmbH 9 1 3 8 Peak Statistics View Popup Command Pointing to Peak Statistics View offers the following commands Figure 9 20 eR y 2D Peak Distribution ROC Curve Single Peak Yariance v Outliers For Box amp Whiskers 2D Options b Figure 9 20 Peak Statistics View submenu Command Used to 2D Peak Distribution Switch to 2D Peak Distribution View and display the 2D peak distribution for two selected peaks ROC Curve Switch to ROC Curve View and display the ROC curve for the selected peak Single Peak Variance Switch to Single Peak Variance View and display the current peak statistics for the selected peak Outliers for Box amp Show Hide the outliers for box amp whiskers plots for the peak Whiskers area intensity per class in the Single Peak Variance View 2D Options Pop up command for displaying data in the 2D Peak Distri bution View 9 1 3 8 1 Peak Statistics View gt 2D Peak Distribution Command The 2D Peak Distribution command switches the Peak Statistics View to 2D Peak Distribution View to display the 2D peak distribution for two selected peaks Shortcut Button amp 9 1 3 8 2 Peak Statistics View gt ROC Curve Command The ROC Curve command is used to switch the Peak Statistics View to ROC Curve View to display the ROC curve for the current pea
92. a small mass shift given in ppm After some postprocessing steps the clustering is converted into a peak list and combined with the initially obtained average peak list Thereby now multiple peaks are ClinProTools User Manual Version 2 2 6 5 Basics Bruker Daltonik GmbH mapped onto a single cluster or peak position Peaks which are very rare in the set of spectra say with a presence of less than 10 can be omitted In that way an overall peak list is obtained which contains peaks which show a nearly overall presence but also rare peaks can be detected which may be present in a single class only The obtained overall average peak list is further processed such that overlapping peaks by means of start end positions but not by means of central masses are made distinct from each other On the obtained list peak features such as the area or intensity are calculated The S N of a peak is determined as an average of the single S N values of the peaks which are mapped to this peak location and can be used for subsequent selection procedures For details please refer to T M Martinez S G Berkovich and K J Schulten Neural gas network for vector quantization and its application to time series prediction IEEE Transactions on Neural Networks 4 pp 558 569 1993 D DeSieno Adding a conscience to competitive learning Proceedings ICNN 88 International Conference on Neural Networks pp 117 124 1988 6 1 1 6 Peak Calculation i
93. able depends on the spectrum s current state Spectra can be excluded or included only before any spectra processing e g recalibration peak calculation is performed In the Spectra and Stack views all excluded spectra are displayed darker colored than the included spectra of the same class e g in dark red instead of light red Figure 9 6 top In the Gel View and the Spectra List report Section 8 1 1 1 manually excluded spectra are highlighted by dark gray bars Figure 9 6 bottom when the Gel Stack View gt Colored Spectrum State command from the View menu is active the automatically excluded spectra are then colored according to the reason of exclusion arb u inProTools ClinProTools Test Data EDTA Run0hiSample_E9_1SLin fid A 100 50 Bi 0 bed 4000 8000 mz Spectrum Number arb u ClinProToolsiC esiClinProToolsic 4000 8000 miz Figure 9 6 Display of a manually excluded spectrum in the Spectra View top the spectrum is indicated by a dark red instead of a light red color in the Gel View bottom it is highlighted with a dark gray bar 9 10 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Reference Part ClinProTools Menus 9 1 2 3 Bitmap to Clipboard Command The Bitmap to Clipboard command defines that a bitmap graphic of the selected data plotting view should be copied to the clipboard when using the Copy command A bitmap graphic is copied with a resolution of 800 600 pixels This format
94. ach individual spectrum or in case of multiple spots on the averages of them 6 4 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Basics 6 1 1 5 1 Peak Picking on the Total Average Spectrum In the total average spectrum peak picking approach standard approach an average peak list is generated by picking peaks on the calculated total average spectrum The automatic detection of peaks is based on the analysis of a smoothed first deri vative The smoothing is determined by the Resolution parameter in the Settings Spectra Preparation dialog Lower resolution values cause stronger smoothing Then the zero crossings of the smoothed derivative are used to identify peaks Internally an iterative procedure using different resolution values is applied to identify unresolved shoulder peaks If the Resolution parameter is chosen too large more and more artificial peaks spikes will be found On the other hand smaller Resolution values will remove more and more unresolved shoulder peaks from the peak list Start and end positions of the peaks are determined by relative slope thresholds of the derivative along the trailing edges of the peak In the total average spectrum peak picking approach standard approach an average peak list is set up by picking peaks on the calculated total average spectrum The number of peaks picked on the total average spectrum and thus the average peak list can be reduced by applying the Signal
95. ack View gt Current Spectrum Marker Command airean treen ees ie esta Seta 9 21 9 1 3 7 3 Gel Stack View gt Colored Spectrum State Command eaire ii etna erent ten eset deke 9 21 9 1 3 7 4 Gel Stack View gt Excluded Spectra Command 9 22 9 1 3 7 5 Gel Stack View gt Group Separators Command 9 23 9 1 3 7 6 Gel Stack View gt Follow Spectra View Mass Range Command eecececceeeeeeeeieeeeeeenieeeeetneeeereae 9 23 9 1 3 8 Peak Statistics View Popup Command cccccceeeceeeeeeeeeees 9 24 9 1 3 8 1 Peak Statistics View gt 2D Peak Distribution Command tees ree eaa i a rene a a Seat tat 9 24 9 1 3 8 2 Peak Statistics View gt ROC Curve Command 9 24 9 1 3 8 3 Peak Statistics View gt Single Peak Variance COMIMANG a a aaa a aa ae a Eear E aa a eSEE TAE 9 25 9 1 3 8 4 Peak Statistics View gt Outliers for Box amp Whiskers Command srei asae ecaehteeias sti A tad aceeets 9 25 9 1 3 8 5 Peak Statistics View gt 2D Options Popup Command aiea a a e aaa a aaa a TAE 9 26 9 1 3 8 5 1 Peak Statistics View gt 2D Options gt Select Peaks Command c08 9 27 9 1 3 8 5 2 Peak Statistics View gt 2D Options gt 95 Confidence Interval Command 9 27 9 1 3 8 5 3 Peak Statistics View gt 2D Options gt Current Spectrum Marker Command 9 28 9 1 3 9 Reset View Settings Command 0 00 0 eect seen eeteeteeeeeenaees 9 29 Data Preparation Menu cccccceeeeeeeceeceece
96. ainst the model There is the same selection done as during model generation by the Noise Spectra Exclusion Adduct Polymer Spectra Exclusion and Similarity Selection It is recommended to use only suitable spectra for external validation In the case of multiple measurements and Similarity Selection switched off the peaks of a group are not averaged but the spectra are treated separately Because of that calculating the recognition capability with the external validation workflow might differ from the one calculated during model generation where the peaks are averaged 6 24 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Basics The model predicts the probable class membership of the validation spectra The Validation report Section 8 1 1 7 includes a so called confusion matrix with one row and one column for each class The entries of the matrix indicate how many spectra from one class have been classified to the correct and to other classes A perfect pre diction would give a diagonal matrix and an average of 100 for the Correct Classified values This view can be used as an indicator for the prediction capability of the model on unknown data and reveals further if some classes are better predicted than others are In addition an individual Classification report Section 8 1 1 8 is set up for each class when the Show Single Classifications option set which shows for each vali dation spectrum assigned to this respect
97. ak calculation and optionally the peak selection parameters in the Settings Peak Calculation and or Settings Peak Selection dialogs as desired 2 From the Data Preparation menu select Peak Calculation 3 To show the corresponding Peak Statistic report select the Peak Statistic com mand from the Reports menu 7 1 7 Manually Excluding Including a Peak After running the peak calculation workflow all picked peaks are indicated by colored integration regions in the Spectra View Included peaks i e peaks that will be used in model generation are indicated by blue integration regions and excluded peaks by gray ones Section 9 1 3 6 6 You can exclude currently included peaks as well as include currently excluded peaks The latter applies to both manually excluded peaks and peaks excluded by automatic peak selection Note Exclusion inclusion of peaks is only possible after the peak calculation workflow was run and if currently no model generation workflow is running To exclude include a peak manually 1 In the Spectra View right click in the integration region of the peak you want to exclude include and select Exclude Peak n resp Include Peak n 7 2 Model Generation and Validation A classification model can be generated by applying one of the four classification algorithms supported by ClinProTools to all included peaks in the non excluded spectra of the loaded model generation classes The resulting model can automatically be vali
98. alculation step has been performed yet To calculate the average peak list 1 Specify the peak picking parameters in the Settings Peak Calculation dialog as desired 2 From the Data Preparation menu select Average Peak List Calculation 7 1 5 2 Manually Editing the Average Peak List The average peak list can be edited manually You can add new peaks to the list change existing peaks with respect to their integration region or remove peaks from the list Moreover a pure manual peak editing is possible as an alternative to automatic peak picking Manual peak editing requires the average peak list calculation workflow to be run first This is needed even if a pure manual peak list editing should be performed For pure manual peak editing average peak list calculation must be run with the Limit Peak Number option in the Settings Peak Calculation dialog Section 9 1 4 2 activated and the Maximal Peak Number set to 0 Editing peaks is also possible after peak statistic calculation or model generation However this resets the current peak calculation indicated by the integration regions of all picked peaks change to gray color and thus requires recalculation of peaks ClinProTools User Manual Version 2 2 7 7 Workflows in Detail Bruker Daltonik GmbH To add a new peak 1 In the Spectra View zoom in the peak you want to add 2 Right click the peak and select Add Peak from the view s context menu This displays the dist
99. altonik GmbH Index Status Bar command 9 12 y Supervised Neural Network ade algorithm 6 12 6 16 Validating mode Supervised Neural Network Cross validation 6 22 parameters 9 47 Externally 6 24 7 15 Support Vector Machine Validation report 7 15 8 9 algorithm 6 12 6 15 Variance command MATLAB 9 84 Support Vector Machine parameters 9 47 Variance for Peak command 9 79 Supporting more than 2 GB RAM 2 3 Variance plot PCA 7 23 System requirements 2 1 Variance window PCA 5 17 View menu 9 11 T View menu MATLAB 9 82 View Spectrum Info command 9 80 Temporary ClinProTools XML files View toolbar 5 10 clearing 4 2 View Toolbar command 9 12 Tool buttons reference A 1 Toolbars W General toolbar 5 10 f Hiding 5 10 Whitewash command 9 80 Showing 5 10 Wilcoxon test 6 27 View toolbar 5 10 Top Hat baseline 6 2 9 32 X Total average spectrum 9 14 Xeaxiscontext m nu 9 70 Total average spectrum calculation 6 4 7 6 XML peak list export format A 13 aN Spectrum pommand a XML spectra import format A 12 U Y Wisdo Zoom command 9 12 Y axis context menu 9 70 Unequal class sizes 6 37 Uninstalling ClinProTools 2 5 Z Unsupervised clustering Zoom command MATLAB 9 83 Calculating i 7 23 Zooming 5 12 Dendrogram window 5 17 Zooming command 9 81 Description 6 36 Performing 7 23 Viewing result 7 24 Unsupervised Clustering command 9 58 Unsupervised Clustering parameters 9 58 ClinProTools User Manual Version 2 2 l 7 Index Bruker Daltonik GmbH l 8 ClinP
100. alySis c cceeececceeeeeeeseeneeeeeeeeeeeeeeeeaeeeeeeaeees 7 19 To i Perorming P GAG isis siccsktde hia EAA EE AEE A indian Gunes eee 7 20 Tot Calculating a PCA na aA crete raedecads leteetdecctiaeeas Gea AA 7 20 amp 5 2 Viewing PCA Results sauciere anA renee edit e AATA 7 21 7 5 2 1 Scores Plots and Loadings Plots 0 0 ceeeseeeeeeeeeeeeeeeeeeeeeneeees 7 21 75 22 Iniuence POT cues ela swan etna 7 22 52 3 Variance Plotis e065 are ne teeter Ra alates 7 23 7 6 Performing Unsupervised CIUStCring ccc ceeeeeeeeeeeneeeeeeenteeeeeeneeeeeeeneeeeetnaeeeeeeaaes 7 23 7 6 1 Calculating an Unsupervised Clustering eeeeeceeeeeeeeeeeteeeeeeneeeteeaeees 7 23 7 6 2 Viewing the Unsupervised Clustering Result 0 ecccceeeeeeeeeeesteeeeeeeeeees 7 24 8 CREPORTING DATA sies cvvessncevtenscs evivescseditvastansesanvestevaveessenetencscteicsensdasted nveitednoeass 8 1 8 1 Creating ClinProtTools Reports eee eceeeneeeeeeeeeee teeter eeeaaeeeeeenaeeeeeeneeeeeeneeeeseaes 8 1 8 1 1 ClinProTools Report Types cccccccececeeececeeeeeeeeeeeeaeceeeeeeesesencaeeeeeeeeeteee 8 2 811 1 Spectra List Report a roet ses taenicl ss feedevidanaiedaite tsa honed teens 8 2 8 1 1 2 Peak Statistic REport ccccceeecssccececeeeeeeeeceeeaeeeeeeesetsnnieaeeeeees 8 3 8 1 1 3 Correlation Matrix Report ceccccecccceceeeeeeeeceeceeeeeeeesesennaeaeeeeees 8 5 8 1 1 4 Correlation List Report 20 0 0 ccececceee
101. alysis ClinProTools offers two algorithms for correlation analysis the standard correlation algorithm and the Kendall s tau b algorithm 6 30 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Basics e Standard correlation algorithm Default algorithm The standard algorithm to determine the correlation matrix list combines ordinary correlation coefficients of pair wise considered peaks in a common matrix e Kendall s tau b algorithm The Kendall s tau b KT algorithm describes a rank correlation coefficient 1 1 It is less frequent than the Spearman rank correlation coefficient a well known alternative algorithm that is not supported by ClinProTools because of the below mentioned reasons however it is much more powerful Within the KT approach all value pairs are compared with each other in a common sense and not just two values of a pair further the error ranks of pairs are evaluated Hence the KT algorithm is less sensitive against outliers KT is more robust and has been recommended if the data do not necessarily come from a bivariate normal distribution Kendall s tau b is a nonparametric measure of association based on the number of concordances and discordances in paired observations Concordance occurs when paired observations vary together and discordance occurs when paired observations vary differently Correlation matrix and per peak correlation list The correlation matrix list is a tool to analyze
102. ance cursor 3 Move the cursor lines to the positions where the peak should start and end and click the right mouse button 4 Confirm the dialog on adding a peak with the given integration region 5 If the selected integration region overlaps with that of an already existing peak confirm the message on peak adding is refused and repeat steps 2 to 4 To change the integration region of a peak 1 In the Spectra View zoom in the peak you want to change 2 Right click the peak and select Edit Peak n from the view s context menu This displays the distance cursor which marks the current integration limits 3 Move the cursor lines to the new positions where the peak should start and end and click the right mouse button 4 Confirm the dialog on changing the peak s integration region as stated To remove a peak 1 In the Spectra View right click the peak you want to remove and select Remove Peak n from the view s context menu 7 1 6 Calculating Peaks and Optionally Selecting Peaks for Model Generation The peak calculation workflow calculates the peaks stored in the average peak list in the single spectra and corresponding peak statistic data as well as selects the peaks to be included in model generation In case of two loaded model classes also the ROC curves per peak Section 6 4 2 2 are generated Peak calculation is based on the peak calculation settings in the Settings Peak Cal culation dialog Section 9 1 4 2 Either t
103. and it will be automatically run before the average peak list calculation workflow starts The picked peaks are indicated in the Spectra View by gray marked integration regions which are shown by default You can cancel the running process by clicking or _Lancel_ 9 1 4 8 Peak Calculation Command The Peak Calculation command is used to run the peak calculation workflow This calculates the peaks stored in the average peak list in the single spectra and specific peak statistic data In case of two loaded model classes also the ROC curves per peak Section 6 4 2 2 are generated Optionally it also performs peak selection for model generation Peak calculation is based on the current peak calulation settings Section 9 1 4 2 Either the peak areas or the maximal peak intensities can be used Peak areas are normalized for model generation when using the GA SVM or SNN Peak selection is performed according to the current peak selection settings Section 9 1 5 1 The command runs peak and peak statistic calculation as well as peak selection if spe cified If the recalibration or the average peak list calculation workflow has not been performed when selecting this command the respecttive workflow s iwill be automatic ally run before the peak calculation workflow starts The result of peak calculation can be viewed in the Peak Statistic report Section 8 1 1 2 Depending on the peak selec tion settings either all peaks or only the selected b
104. arameters are set in the Settings Spectra Prepara tion Section 9 1 4 1 and Settings Peak Calculation Section 9 1 4 2 dialogs 6 1 1 Standard Data Preparation Workflow The spectra selected for model generation and classification are treated according to a standard workflow generally including the following steps e Baseline subtraction on spectra e Normalization of spectra e Recalibration of spectra optional e Average spectra calculation e Average peak list calculation ClinProTools User Manual Version 2 2 6 1 Basics Bruker Daltonik GmbH e Peak calculation in the individual spectra e Normalization of peak lists for model generation 6 1 1 1 Baseline Subtraction on Spectra The purpose of the baseline subtraction is to remove the broad structures of a spec trum If we would not do a baseline correction the variable level of the baseline which depends on the preparation would influence the peak areas quite a lot and would make it difficult to select peaks based on S N and intensity thresholds The baseline subtraction is done 1 on the individual spectra to prepare the spectra for the purpose of recalibration and quality checks and 2 again on the average spectrum The latter baseline correction is done to remove baseline structures which come up from the averaging of the noise of the individual spectra The goal of the baseline algorithms is to remove the broad baseline structures without disturbing the line shape
105. arted automatically within ClinProTools 7 6 1 Calculating an Unsupervised Clustering An unsupervised clustering is calculated on all non excluded spectra in the loaded spectra set s and requires three valid spectra with three peaks being available at least The unsupervised clustering automatically runs the spectra recalibration average peak list calculation and or peak calculation workflows if these have still not been performed when launching unsupervised clustering calculation After the unsupervised clustering is completed the Dendrogram window opens displaying the created dendrogram To calculate an unsupervised clustering 1 Open the spectra set s you want to cluster ClinProTools User Manual Version 2 2 7 23 Workflows in Detail Bruker Daltonik GmbH 2 If certain spectra should not be included in unsupervised clustering exclude them 3 From the Statistical Analysis menu select Unsupervised Clustering or click 4 In the Unsupervised Clustering dialog specify the parameters as desired and click OK to start unsupervised clustering If required the spectra recalibration average peak list calculation and or peak calculation workflows will be run prior to starting unsupervised clustering 5 View the resulting dendrogram 7 6 2 Viewing the Unsupervised Clustering Result The result of an unsupervised hierarchical clustering of spectra can be viewed in the Dendrogram window Figure 5 17 The created dendrogram show
106. aseline Convex Hull Baseline 10 0 a Minimal Baseline Width Baseline Flatness Mass Range m z 0 100000 Minimal Mass Maximal Mass Savitsky Golay Smoothing Enable lt Width m z Cycles Data Reduction Enable Factor Figure 9 29 Cancel Help Resolution Defaults Null Spectra Exclusion V Enable Noise Spectra Exclusion Enable Noise Threshold Adduct Polymer Spectra Exclusion Advanced Enable i fs Spectra Grouping Support Spectra Grouping 4 Recalibration V Enable 1000 ppm Maximal Peak Shift 30 Match to Calibrant Peaks IV Exclude not Recalibratable Spectra Settings Spectra Preparation dialog default setting ClinProTools User Manual Version 2 2 9 31 Reference Part ClinProTools Menus Bruker Daltonik GmbH In Resolution define the resolution Section 6 1 3 1 to be applied to the peak detec tion algorithm as a hint for the peak width Resolution Enter the resolution to be applied to detecting peaks for individual spectra used for recalibration and peaks for the total average spectrum used for model building If this parameter is chosen too large more and more artificial peaks spikes will be found On the other hand smaller resolution values will remove more and more unresolved shoulder peaks from the peak list Alternatively you can select the respective mass range from the drop down list belo
107. ass gt List of spectra paths per class lt SpectraPaths gt lt Areas gt lt Class gt List of peak lists areas per class lt Areas gt lt Intensities gt lt Class gt List of peak lists intensities per class lt Intensities gt lt ClinProToolsPeakLists gt The XML2 Files format generates an alternative XML format lt ClinProToolsPeak Lists2 gt with the class and spectra paths provided as attributes ClinProTools User Manual Version 2 2 A 13 Appendix Bruker Daltonik GmbH The XML3 Files format lt ClinProToolsPeakLists3 gt is similar to XML2 Files but a style sheet reference for ClinProtPeakList xsl is added to facilitate working with peak lists in Excel A 5 Part Numbers 249614 Software Package ClinProTools 2 2 249620 License ClinProTools 2 2 245575 License Support Vector Machine 1 0 249619 ClinProTools User Manual A 14 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Index 1 B 1D peak distribution 9 17 Background Color command 9 71 Baseline subtraction 6 2 2 Baseline subtraction filter 6 2 9 32 Basic ClinProTools workflows 4 3 2D Options popup command 9 26 Batch classification 6 25 2D peak distribution 5 6 9 24 Bitmap to Clipboard command 9 11 2D Peak Distribution command 9 24 Box amp Whiskers command 9 18 2D Peak Distribution View 5 6 Box and whiskers 5 8 2D Peak Distribution View context Browse ClinProTools Folder command 9 6 menu 9 68 C 9 l Calculate command 9 50
108. ass 2 sensitivity and specificity can be derived by considering the confu sion matrix in the Validation report as follows Figure 6 9 ClinProt Validation Correct Classified cl N 1 2 0 Inv a ame Part of Valid Spectra i 1 Diseased 88 9 24 1 303 0 0 2 Control 73 1 72 19 4 0 0 Figure 6 9 Validation report used in our example 6 42 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Basics From the confusion matrix which in our example includes the columns labeled with 1 classifications to class 1 2 classifications to class 2 0 unclassified spectra and Inv number of invalid spectra count for Class 1 diseased 1 the number of correct positive classified spectra as TP true positives 2 the number of wrong positive classified spectra as FP false positives Class 2 control 3 the number of wrong negative classified spectra as FN false negatives 4 the number of correct negative classified spectra as TN true negatives Thereby the rows can be seen as the true classifications taken from the sample set and the columns indicate the prediction of the machine including not classifiable samples Considering the example of the above Validation report for the validation of two classes class 1 diseased class 2 control one obtains Sensitivity The sensitivity of such a test is the probability that the test has a positive outcome when the te
109. ass shift allowed for a peak in recalibration in ppm Values from 1 to 2000 ppm can be set Match to Calibrant Peaks Enter the percentage match to calibrant peaks value which is multiplied with the Maxi mum Quality Value number of reference masses to determine the Spectra Quality Threshold of the filter The spectrum s Spectrum Quality Value must reach this thres hold so that the spectrum is not marked as not recalibratable 0 means no exclusion the highest reasonable value probably is 80 Exclude Not Recalibratable Spectra Check this option if spectra that are marked as not recalibratable should be excluded in further processing OK Changes the spectra preparation settings Depending on the current processing state the views may become cleared to prevent the loaded spectra from further processing then a message will inform you on how to proceed 9 36 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Reference Part ClinProTools Menus 9 1 4 2 Settings Peak Calculation Command The Settings Peak Calculation command is used to set the parameters for picking peaks and calculating peak areas intensities and statistical data The settings are stored with the spectra preparation settings in the SettingsDataPreparation xml The command opens the Settings Peak Calculation dialog Figure 9 32 Note The peak calculation settings especially the Signal to Noise Threshold may strongly influence the quality of
110. ation Regions command shows hides the integration regions of the picked peaks in the Spectra View Figure 9 10 The integration regions are highlighted with different colors concerning the current state of the peak Non excluded peaks are indicated in blue and excluded ones in gray Peaks incorporated in the selected calcu lated model are marked red Peaks forced into a model are green highlighted before and after model generation as well The integration regions are shown by default arb u Files ClinProTools ClinProTools Test Data Spiked Data Normal0_M23_1SLin fid zi 80 60 40 20 1200 1300 1400 miz Figure 9 10 Coloring of integration regions of peaks that are included blue excluded gray forced into the model green and incorporated in the model red Shortcut Button ak 9 1 3 6 7 Spectra View gt Average amp StdDev Command The Average amp StdDev command shows hides the per class average with standard deviation plots in the Spectra View Figure 9 11 These represent the calculated aver age of the peaks areas intensities in the single spectra of a class with the correspond ing standard deviation on both sides These bars are colored like the corresponding class The plot is drawn on a unique scale independent of the peak intensity scale These bars are hidden by default Shortcut Button JE 9 16 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Reference Part ClinProTools
111. ault setting Clicking OK runs a PCA on the non excluded spectra in the data set s If the spectra recalibration average peak list calculation and or peak calculation workflows have not been performed yet the respective workflow s will be automatically run before starting PCA calculation If the Check Memory for PCA option in the General Settings dialog is set first the available memory is checked if it is sufficient for PCA on the loaded data set s After the PCA is completed the PCA main window opens displaying the results of the PCA in the Scores and Loadings plots Section 7 5 2 ClinProTools User Manual Version 2 2 9 57 Reference Part ClinProTools Menus Bruker Daltonik GmbH Shortcut Button 9 1 7 2 Unsupervised Clustering Command The Unsupervised Clustering command is used to run the unsupervised clustering workflow performing an unsupervised hierarchical clustering Section 6 4 2 4 on the non excluded spectra of the loaded spectra data set s This calculates clusters from the spectra classes and creates a dendrogram showing the distances among the single clusters The corresponding data is stored in the files ClinProtClustering xml ClinProtClusteringTree xml and ClinProtClusteringTree2 xml in the ClinProTools folder An unsupervised clustering needs three valid spectra with tree peaks being available at least The command can be applied to several classes or to only a single class However in the context of
112. be applied onto one anchor position of the target and allowed to dry at room temperature It is recommended to work continuously as matrix and several sample solutions contain very volatile solvents uncontrolled evaporation may result in decreased preparation quality The measurement variance is reduced by spotting each sample several times Matrix for the mass range 5 100 kDa Matrix solution 7 6 mg 50 umol 2 5 DHAP are suspended in 375 ul EtOH and 125 ul 10 umol of diammonium hydrogen citrate stock solution 27 mg in 1 5 ml distilled H20 are added The suspension should be vortexed for at least 1 min followed by sonification for 15 min The mixture has to be vortexed again 1 min and the clear matrix solution is suitable for MALDI TOF MS analysis now For target preparation 2 ul of sample are acidified with 2 ul of 2 TFA Subsequently 2 ul of freshly prepared 2 5 DHAP matrix solution have to be added and vigorously mixed Finally 1 ul of the mixture should be applied onto one anchor position of the target and allow to dry at room temperature Parallel spotting on multiple target posi tions is recommended as well 3 3 Data Acquisition with flexControl A Bruker MALDI TOF mass spectrometer will be delivered with a number of acquisition methods which were specifically adapted to the individual machine during installation and which can be loaded directly into the acquisition software flexControl Those default methods cover e g
113. calculate a so called One way ANOVA A One Way Analysis of Variance is a way to test the equality of three or more means at one time by using variances Assumptions e The populations from which the samples were obtained must be normally or approxi mately normally distributed e The samples must be independent e The variances of the populations must be equal The null hypothesis will be that all population means are equal the alternative hypo thesis is that at least one mean is different If the decision is to reject the null then at least one of the means is different However the ANOVA does not tell you where the difference lies 6 4 1 3 Wilcoxon Test The Wilcoxon rank sum test is a non parametric alternative to the paired Student s t test This test should be used whenever the assumptions that underlie the t test cannot be satisfied The test is named for Frank Wilcoxon who proposed this and the rank sum test in 1945 The null hypothesis tested is that a sample is symmetrically distributed around a speci fied center It is often used to test difference scores of data collected before and after an experimental manipulation in which case the central point would be expected to be zero Scores exactly equal to the central point are excluded and the absolute values of ClinProTools User Manual Version 2 2 6 27 Basics Bruker Daltonik GmbH the deviations from the central point of the remaining scores are ranked such tha
114. ceeeceeeeceeeeeeeeeeeeccnceeeeeeeesetennieaeeeeeees 9 4 9 1 1 5 Info Loaded Classes COMMANG ccccceeeeseceeeeeeeeeeteetteaeeeeees 9 4 9 1 1 6 Save Class Paths Command cccccceceeeeeececeeeeeeeeseeeenteeeeeeees 9 5 QTE Print Command sssri aoe ai aie die eee 9 5 9 1 1 8 Print Preview Command cecceeeccececeeeeeeeeceeeeeeeeeeeseteenieaeeeeees 9 5 9 1 1 9 Print Setup Command ccccceceeeece cece eee eeeeceneeeeeeeeeseteenieaeeeeeess 9 6 9 1 1 10 Peak List Export Command ccceceeeeeeeeeeeeeeeeeeesetennieaeeeeees 9 6 9 1 1 11 Browse ClinProTools Folder Command ccccccseeeeseceeeteees 9 6 9 1 1 12 General Settings COMMANG cece eeeeeeeeteteeeeeteteeeeeeneeeerene 9 6 9161213 Exit Command en a Rade rie eaters ated reel 9 9 9 127 JEdit MENUn iei iaae aaa a a a a a aaa a aaa aitaa 9 9 9 1 2 1 Coy COmmMman aiaa ar a cada R E eet 9 9 9 1 2 2 Exclude Include Spectrum Commanid ccccceceeeeeceeeeeees 9 10 9 1 2 3 Bitmap to Clipboard Command cccceceeceeceeeeeeeeeeeeeeeeeeees 9 11 9 1 2 4 Metafile to Clipboard Command cccccccceeeeeeeeeeseettceeeeeeees 9 11 9 1 37 View MON Uiesita ioietan che ae aee aa e lanstesants Hat eaaa aaa E denceetan tabs 9 11 9 1 3 1 General Toolbar Commangd ccccceceeeeeeceeceeeeeeeteeeennnaeeeeees 9 12 9 1 3 2 View Toolbar COMMANG cccccceececeeeeeeeeececaeeeeeeesete
115. ch E ag 4 Cee ms E 2200 2300 t2 e Data partitions 9 one K prototypes e Prototypes Ld jaie peak eg crossover points iig Class 3 e Peak ranking combinations Best separating i Class 1 Evolutionary characteristics peak combination Data distribution characteristics 1700 1800 2000 Figure 6 1 Overview of ClinProTools four classification algorithms 6 2 1 1 Genetic Algorithm The concept of Genetic Algorithms GA was developed by John Holland J H Holland Adaptation in Natural and Artificial Systems University of Michigan Press Ann Arbor 1975 It is based on the idea of evolution in which the fittest individuals have the highest chances of survival Here we apply them to select combinations of peaks which perform best in separating the classes under consideration Pattern determination is used to identify an optimal set of peaks which gives the best separating model determined upon the model generation spectra used and validated on test spectra or by a cross validation procedure A brute force approach would not work A systematic trial of all combinations would take far too long because the number of possible combinations is extremely large For 1000 given peaks and a desired com bination of just 3 peaks you get 1 000 999 998 997 002 000 sets of peaks There fore we need more sophisticated ways to do it ClinProTools User Manual Version 2 2 6 13 Basics Bruker Daltonik GmbH The advantage of
116. ch effects are observed one should take a closer look on the values for this feature mass by analyzing the exported peak list The peak list can be exported to XML or CART format Section 8 4 From the above explanations it becomes obvious that the more flexible the constraints 2 2 classes distribution free small sample size large number of features the more complicated the test scenario For very relaxed constraints most tests are not powerful enough and the obtained results are in fact invalid To overcome this one has to except some constraints to make the problem more suitable For example one could increase the number of disjunctive samples by 10 or 100 if possible this will improve correct ness of the underlying estimations and improve the performance of the tests It is also very common to reduce the number of features to a smaller subset e g 100 features by omitting features which are probably unimportant e g because of some pre know ledge If one ignores these problems or the data just does not fit to these aspects the obtained p values from the statistical tests are in fact poorly estimated and may be inappro priate If e g the number of samples is small and the normal distribution assumption is not true p values from the t test or the ANOVA test may be very unrealistic small To take now just the alternative distribution free test is also no solution hence this test requires a much larger number of samples to obtain the
117. cient The resulting mass devia tion over the whole target is mostly lt 300 ppm If Prespotted AnchorChip targets are used nearest neighbor calibration should be done to prevent higher mass deviation In practice 2000 ppm is an upper estimate for the individual mass error The task of recalibration is to reduce mass shifts occurred during the measurement Spectra recalibration is enabled by default but can be turned off In order to recalibrate single spectra a list of reference masses is required Such a list is obtained from the line spectra derived from the original data using peak picking Only those masses which occur in at least 30 of the spectra are used as reference masses The recalibration algorithm looks for these reference masses in the peak list of each spectrum The ppm Maximal Peak Shift parameter of the Settings Spectra Pre paration dialog is used as upper limit of the mass difference between reference mass and peak mass The calibration function of the spectrum is modified such that the mass error across all assigned pairs is minimized The number of reference masses depends on the application For measurements within the mass range 1 000 10 000 Da about 40 80 reference masses can be expected Typically 50 and more of those peaks are used for the recalibration of individual spectra As part of the recalibration step the list of reference masses is generated After recali bration it is checked by the spectrum quality f
118. cified name Reset current model generation settings to their defaults Classify the spectra in the selected collection with chosen model Validate the selected model externally using test spectra for each class Save the current classification result in an XML file with a specified name Button Shortcut Calculate i Cancel Load Clear All Hi Classify Validate il A 4 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Appendix Menu commands Used to Button Shortcut Show Classification Show the classification result for the classified spectra in the Classification report Close Classification Close the current classification eT Statistical Analysis menu PCA Perform PCA on the spectra data JE set s Unsupervised Clustering Perform hierarchical clustering on the f spectra data set s Reports menu Spectra List Create and show the Spectra List report Peak Statistic Create and show the Peak Statistic 72 report Correlation Matrix Create and show the Correlation Matrix report Model List Create and show the Model List report Model List Settings Statistic Define settings for calculating peak statistic and showing certain statistical data in the Spectra View Compass menu LicenseManager Launch Bruker Daltonics LicenseManager Help menu Help Topics Launch ClinProTools online help F1 About ClinProTools Display copyright and license information about the present ClinP
119. cision is not possible for more than two classes Within Figure 6 6 the ROC curve is explained on an example of two populations diseased and non diseased patients For all patients the same test is performed and numeric results are received for each patient A plot of these results leads to the upper diagram shown The vertical green line within the diagram indicates an arbitrary chosen threshold a value above this threshold represents a positive test result and a value below it a negative test result The position of this cut off point will determine the number of true positives TP true negatives TN false positives FP and false negatives FN If the test threshold is moved from left to right the proportion of the FP decreases but the TP also decrease simultaneously lower diagram The ROC curve graph right to the lower diagram is an exploration of what happens to the TPF and the FPF if the position of the arbitrary threshold is varied The point corresponding to the chosen threshold is shown on the ROC curve as the cross If the threshold is very high almost no FP occur on the one hand but only less TP are identified on the other hand If the threshold is moved towards a more reasonable lower value the number of TP increases the ROC curve moves steeply up Finally a region will be reached where there is a remarkable increase in FP and the ROC curve slopes off as the test threshold is moved down to ridiculously low values Th
120. classification algorithm and the cross validation settings The command calculates all added state models present in the list at once If the recalibration average peak list calculation or peak calculation workflow has not been performed when selecting this command the respective workflow s will be automatically run before model calculation starts After model generation is completed the corresponding model data is entered in the model list with changing the state of the model s into Calculated Shortcut Button Calculate 9 1 5 4 Cancel Command The Cancel command is used to cancel any currently running spectra loading recali bration peak calculation model generation or classification process Same as Cancel command from File menu 9 1 5 5 Load Model Command The Load Model command is used to load the XML file of a model that has previously been saved with a specified name Section 9 2 9 19 This allows performing classifi cation or external validation The command opens the Load Model dialog with the ClinProtModels folder opened by default Navigate to the model you want to load and click Open This enters the model in the model list with the state Loaded Shortcut Button Load 9 1 5 6 Clear All Command The Clear All command is used to clear the Model List View This removes all items currently present Note ClinProTools does not save models automatically Thus before selecting this command you should
121. classified in the model according to the data preparation and model generation parameters stored in the model ClinProTools supports two modes for spectra classification launching different workflows 7 3 1 Changing the Classification Mode Spectra classification can be run in standard or in batch mode Section 6 3 with the standard mode being active by default If you want to run your next spectra classifica tion in another mode than that currently active you have to change the mode before starting the classification Changing the mode is not possible when a classification is running or loaded To change the current classification mode 1 From the File menu select General Settings 2 In the Settings General dialog set Classify in Batch Mode as needed Check the option to work in batch mode or if the mode has been previously switched to batch mode uncheck the option to work in standard mode again Then click OK 7 3 2 Selecting a Model for Spectra Classification To classify spectra you have to select the model to be used in the model list The model should be suitable for the data you want to analyze To select the model to use 1 In the model list select the model you want to use If the desired model is currently not in the model list load it 7 16 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Workflows in Detail 7 3 3 Selecting the Spectra to be Classified and Running Classification After selectin
122. commands for showing data in the Peak Statistics View Reset View Settings Reset certain settings of the data plotting views to the defaults 9 1 3 1 General Toolbar Command The General Toolbar command shows hides the General toolbar The General toolbar is shown by default 9 1 3 2 View Toolbar Command The View Toolbar command shows hides the View toolbar The View toolbar is shown by default 9 1 3 3 Status Bar Command The Status Bar command shows hides the status bar The status bar is shown by default 9 1 3 4 Undo Zoom Command ClinProTools stacks the zooming operations you perform in the Spectra Gel 2D Peak Distribution or Single Peak Variance View for each view separately The Undo Zoom command undoes the last change in zoom range performed in the currently focused view Shortcut Button Q 9 12 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Reference Part ClinProTools Menus 9 1 3 5 Redo Zoom Command The Redo Zoom command restores the previously undone zoom range in the currently focused view Shortcut Button a 9 1 3 6 Spectra View Popup Command Pointing to Spectra View offers the following commands Figure 9 8 Spectra View d v Single Spectra All Single Spectra Total Average Spectrum Average Spectra Noise Spectrum v Integration Regions Average amp StdDey Peak Distribution Box amp Whiskers v Outliers For Box amp Whiskers Peak Markers Figure 9 8 Spectra View
123. coxon Kruskal Wallis Figure 9 34 Settings Peak Selection dialog default setting Use All Check this option if you want to use all picked peaks in model generation Uncheck it if you want to restrict the number of peaks to a maximal number of best ones Peaks To Use If Use All is unchecked enter the maximal number of best peaks to use The peaks are selected with respect to the Sort Mode Sort Mode Select how to sort peaks if only the best ones should be used Difference Average Sorts the peaks by the difference between the maximal and the ClinProTools User Manual Version 2 2 9 43 Reference Part ClinProTools Menus Bruker Daltonik GmbH minimal average peak area of all classes P Value T Test ANOVA Sorts the peaks by the p value from t test Section 6 4 1 1 ANOVA test Section 6 4 1 2 P Value Wilcoxon Kruskal Wallis Sorts the peaks by the p value from Wilcoxon test Section 6 4 1 3 Kruskal Wallis test Section 6 4 1 4 OK Changes the current peak selection settings If peak selection has already been per formed the current selection is changed according to the new settings 9 1 5 2 New Model Command The New Model command is used to add a new model parameter set to the model list This launches selecting the classification algorithm to use specifying the algorithm specific parameters and entering a model name The command opens the Choose Algorithm dialog Figure 9 35 to select the classifycation a
124. ctra Import XML File cceeceeeeeeeeeeeeeeeeeeeeeeees 7 5 7 1 3 Manually Excluding Including a Spectrum ecceeccncceeeeeteeeeeeetieeeeettieeeeeeee 7 5 7 1 4 Recalibrating Spectra and Calculating Average Spectra ee eeeeeeee 7 6 7 1 5 Setting up the Average Peak List eceeeeceteeeeeeenneeeeeeeteeeeetteeeeeenaeeeees 7 7 7 1 5 1 Calculating the Average Peak List 0 ecceeeeeeeeteteeeeeenneeeeeees 7 7 7 1 5 2 Manually Editing the Average Peak List 0 eccceeeteeetteeeeeeeees 7 7 7 1 6 Calculating Peaks and Optionally Selecting Peaks for Model Generation 7 8 7 1 7 Manually Excluding Including a Peak cceceeeeeeeieeeeeeeneeeeeetieeeeetnieeeereea 7 9 7 2 Model Generation and Validation cceeececceeeeeeeeeeeenneeeeeeaeeeeeeaeeeeeenaeeeeeeetaeeeeeeaas 7 9 O21 Generating aiModel ceca E EE AE atid eine 7 10 7 2 1 1 Defining the Model Generation Settings ee cceeeeeteeeeeeeees 7 10 ClinProTools User Manual Version 2 2 v Contents Bruker Daltonik GmbH 7 2 1 1 1 Adding a Model Parameter Set to the Model List 7 10 7 2 1 1 2 Setting the Cross Validation Parameters 0 7 11 7 2 1 1 3 Saving Loading and Resetting the Model Generation Settings cccceeeeeeeeeeeeeeeeeeeeteeeeeenaeees 7 11 7 2 1 2 Checking and Optionally Changing the Current Peak Selection 7 12 7 2 1 3 Forcing a Peak into a Model cc eeeeeeeeeeeeeeeeeeeeeeeeeeneeeeeenaee
125. d Single Peak Variance View Single Peak Variance View context menu Single Spectra command Small p value phenomenon Smoothing filter 6 9 Specificity 6 10 9 40 9 53 7 17 7 3 7 14 7 11 8 12 9 5 6 9 9 78 6 35 7 21 9 27 6 42 Spectra classification 4 6 Spectra filtering 6 8 Spectra grouping 6 7 Spectra import XML file Opening 7 5 Saving 9 5 Spectra import XML format A 12 Spectra List command 9 61 Spectra List report 8 2 Spectra loading 7 4 Spectra normalization 6 3 Spectra preparation parameters 7 1 9 30 Spectra quality filter 6 11 9 36 Spectra recalibration 6 3 Spectra View 5 2 Spectra View context menu 9 67 Spectra View popup command 9 13 Spectrum Calculating peaks 6 6 7 8 Classifying 4 6 6 25 7 16 Closing 9 4 Distance measurement 9 73 Excluding 7 5 Including 7 5 Loading 7 4 Normalizing 6 3 Recalibrating 6 3 7 6 Subtracting baseline 6 2 Spectrum marker 2D Peak Distribution View 9 28 Gel View 9 21 Stack View 5 5 Stack View context menu 9 68 Stack View orientation change 5 13 Standard box amp whiskers plot 9 18 Standard classification 6 25 Standard correlation algorithm 6 31 Standard deviation 9 27 Starting ClinProTools 4 1 Statistical Analysis menu 9 57 Statistical methods 6 30 Statistical problems with MS data 6 37 Statistical tests 6 26 Status bar Description 5 10 Display of coordinates 5 10 Hiding 5 10 Showing 5 10 l 6 ClinProTools User Manual Version 2 2 Bruker D
126. d however displaying big classifications is not recommended because the browser used for display might take a very long time to process the XML file with style sheet Showing the classification result is possible as long as you do not close the current classification Section 7 3 6 To show the classification result 1 From the Classification menu select Show Classification 7 3 6 Closing the Classification Closing a classification removes the current classification result from the memory and in the standard mode it also unloads the classified spectra and removes them from the views To close the classification 1 From the Classification menu select Close Classification or click a N 7 4 Peak Statistic and Correlation Analysis Calculation 7 4 4 Calculating Peak Statistic The peak statistic calculation workflow is in principle the same like the peak calculation workflow but additionally creates and shows the Peak Statistic report Section 8 1 1 2 That report lists all peaks picked in the spectra of the loaded model generation classes with corresponding state peak area intensity and statistical data By default the peak statistic workflow uses the current peak selection settings Section 9 1 5 1 defined for 7 18 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Workflows in Detail model generation but if desired you can define different settings for reporting For example you can sort peaks by m z value a
127. d average and normalized to 1 If the value is 1 for all classes all classes are equally likely If the value is below 1 the class ClinProTools User Manual Version 2 2 6 19 Basics Bruker Daltonik GmbH is less likely if it is higher it is more likely The class with the highest likeliness is the predicted one 6 2 1 5 Detection Modes to Determine the Best Number of Peaks in a Model Note The peak rankings in the multivariate algorithms GA SVM SNN are derived from an analysis in a high dimensional data space performed on all peaks passed to the algorithm These rankings may underestimate the individual importance of single peaks as available by use of univariate rankings obtained from statistical tests t test ANOVA in some cases they might differ signifi cantly Especially if the number of peaks is limited or fixed which is an option in GA and SVM it might happen that the algorithms choose peaks with a subopti mal univariate classification capability If a classification based on a univariate peak ranking is desired it is advisable to pre sort the peaks with the Peak Selection available in the Settings Peak Selection dialog in the Model Generation menu E g the peaks used for the algorithms can be reduced to the three best peaks according t test In this way the algorithms are forced to use the best peaks according to a univariate ranking ClinProTools offers an automatic and a manual detection mode for determin
128. d to generate the model and then are classified in the model All settings for the respective workflows are stored in the model From the confusion matrix in the Validation report Section 8 1 1 7 you can obtain how well your current model classifies the known test spectra To perform an external validation on a model 1 In the model list select the model you want to validate If the desired model is currently not in the model list load it 2 From the Classification menu or the Model List View context menu select External Validation or click Validate 3 In the External Validation dialog Select for each class present an appropriate spectra collection to be classified Enter the path and name of the folder containing the collection or click Browse and navigate to the respective folder ClinProTools User Manual Version 2 2 7 15 Workflows in Detail Bruker Daltonik GmbH Specify whether single classification reports Section 8 1 1 8 per class should be shown 4 Click OK to start the external validation workflow This prepares and classifies the validation spectra and then shows the Validation report and the single Classifica tion reports if created 7 3 Spectra Classification Within ClinProTools unknown spectra can be classified in a classification model set up for the respective analytical task The spectra to be classified are loaded and prepared as the spectra that were used for model generation and are
129. dated internally by cross validation within the model generation workflow An exter nal validation can also be performed after model generation ClinProTools User Manual Version 2 2 7 9 Workflows in Detail Bruker Daltonik GmbH 7 2 1 Generating a Model You can generate a new model when at least two model generation classes are loaded and the data preparation has already been done or you have defined the settings you want to use for it To generate a new model you have to add a new model parameter set to the model list that defines the classification algorithm to be used and the algo rithm related model parameters The peaks that should be included in model genera tion have to be determined as well as the settings for cross validation before model calculation is started There is no fixed order in which these individual steps have to be done the following description will start with adding a new model parameter set 7 2 1 1 Defining the Model Generation Settings The model generation settings define how model generation validation and spectra classification is performed They are specified in the Settings Peak Selection Section 9 1 5 1 Settings Genetic Algorithm Section 9 1 5 2 1 Settings Support Vector Machine Section 9 1 5 2 2 Settings Supervised Neural Network Section 9 1 5 2 3 Settings QuickClassifier Section 9 1 5 2 4 and Settings Cross Validation Section 9 1 5 7 dialogs You can use the default parameters or speci
130. ded spectra in the Gel and Stack views Excluded spectra are shown by default 9 22 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Reference Part ClinProTools Menus 9 1 3 7 5 Gel Stack View gt Group Separators Command The Group Separators command shows hides group separators in the Gel View Figure 9 19 Group separators are horizontal dashed lines which separate spectra groups resulting from multiple measurements of samples within one class These markers are only displayed if the Support Spectra Grouping option in the Settings Spectra Preparation dialog is checked on spectra loading Group separators are shown by default 2000 4000 6000 8000 miz Figure 9 19 Separation of groups of spectra from multiple measurements by horizon tal group separators here each group consists of four spectra 9 1 3 7 6 Gel Stack View gt Follow Spectra View Mass Range Command The Follow Spectra View Mass Range command force or do not forces the Gel Stack View to follow the mass range of the Spectra View By default the x axis of the Spectra View and the x axis of the Gel Stack View show a slave master behavior The x axis of the Spectra View always follows the x axis of the Gel Stack View when the latter is changed but contrarily the x axis of the Gel Stack View is kept when the x axis in the Spectra View is changed Enabling the command switches the x axis of the Gel Stack View to dependence on the x axis of the Spectra
131. dialog ClinProTools User Manual Version 2 2 9 85 Reference Part MATLAB Based Menus Bruker Daltonik GmbH 9 86 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Error Treatment 10 ERROR TREATMENT Error reports In the case that an error occurs please send an error report to clinprot support bdal de The error report should contain the following e The dump file dmp from C BDALSystemData Dumpfiles if a message box has appeared with a dump file has been generated e If a message box titled ACO or SCO pops up with an error the log file C BDALSystemData BFADSLOG TXT an additional screen shot is also good e The ClinProTools build number from the About box e A short description of the workflow e Is the error reproducible if yes how e The data resp which kind of data have been used e The settings files SettingsDataPreparation xml SettingsModelGeneration xml und SettingsGeneral xml from the ClinProTools folder e In the case of an ERROR entry in the Model List View select the Show Error command from the view s context menu the Internet Explorer pops up and the file can be saved with the Explorer s Save As command Reset of ClinProTools in case of error message After encountering an error message the ClinProTools software could be in an intermediate state and it might be necessary to reset the software using the Close All command from the File menu External shutdown of ClinProTools i
132. ds the selected resp all models present Note Please remember that models are not saved automatically in ClinProTools Thus if you have calculated a new model you should first consider whether to save it Section 7 2 1 7 before removing it 7 14 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Workflows in Detail To remove a single model from the model list 1 In the Model List View right click the model you want to remove and select Remove Model To clear the model list 1 From the Model Generation menu select Clear All or click _ Clear All 7 2 1 9 Loading a Model A model that was saved in a XML file can be loaded in the model list again e g to perform external validation on it or classify unknown test spectra in it A loaded model gets the state Loaded To load a model 1 From the Model Generation menu select Load Model or click td This opens the Load Model dialog with the ClinProtModels folder opened by default 2 Navigate to the model you want to load Double click it or select it and click Open This enters the model in the Model List View 7 2 2 Validating a Model Externally You can perform an external validation on a calculated model using spectra of known class membership that have not been used in generating the respective model For each class in the model corresponding spectra must be loaded The validation test spectra are loaded and prepared like the model generation spectra use
133. e The identification of biomarker candidates within ClinProTools focuses on the detected peaks over a given set of spectra If the number of detected peaks is large we also have a large number of features peak areas Applying statistical tests on each feature at the same time forms the case of the so called multiple hypothesis testing The application of a statistical test on a single feature aims on single hypothesis testing This is the case when we want to know if the given feature derived from a peak shows a significant difference between the considered classes This is a single hypothesis In general however the number of features is large and for each feature we create a hypothesis which is tested by some statistical test If we do so we are considering multiple hypotheses Testing each of a large number of hypotheses at the same alpha level as for a single hypothesis normally leads to a large number of false positives i e features are called significant although there is no expression change in reality To overcome this problem different p value adjustment procedures have been proposed in the statistical literature ClinProTools User Manual Version 2 2 6 41 Basics Bruker Daltonik GmbH ClinProTools automatically applies the so called Benjamini amp Hochberg p value adjust ment procedure to adjust the p values to observe the multiple hypothesis problem For details we refer to S Dudoit and J Popper Shaffer and J
134. e ig e OF e Figure 9 49 Settings Statistics dialog default setting ClinProTools User Manual Version 2 2 9 63 Reference Part ClinProTools Menus Bruker Daltonik GmbH Use Selection Sort Mode from the Settings Peak Selection Dialog Check this option if the selection sort mode defined in the Settings Peak Selection dialog Section 9 1 5 1 should also be used in peak statistic Otherwise uncheck this option and specify parameters as desired In Peaks to Show in Views define whether statistical data in the Spectra and Single Peak Variance views should be displayed for all peaks or only a specified number of best peaks with respect to the selected sort mode Show All Check this option if statistical data should be shown for all peaks Uncheck it if you want to limit the number of peaks Peaks to Show If Show All is not checked enter the maximal number of peaks for which statistical data should be shown The selected Sort Mode determines the order of peaks In Sort Mode define how to sort the peaks in the Peak Statistic report if the current sort mode from peak selection should not be used In addition this setting defines the selection of peaks for which statistical data should be shown in the Spectra View if the peak number is limited Mass Sorts by the m z value Difference Average Sorts the peaks by the difference between the maximal and the minimal average peak area of all classes P Value T Test AN
135. e the peak area resp the intensity data of the peak list can be exported to the CART ASCII format dat by Salford Systems San Diego CA USA Appen dix A 4 For each spectrum the class membership and the areas resp intensities of the picked peaks given by their m z value are exported To export the peak list to XML or CART format 1 From the File menu select Peak List Export This opens the Peak List Export dialog 2 Navigate to the folder where you want to save the exported file 3 Enter a name for the exported file or select one from the folder list 4 In Files of Type select the desired format XML Files XML2 Files XML3 Files or CART Files 5 Click Save If you have chosen an existing file name confirm the appearing message to overwrite the file 8 14 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Reference Part ClinProTools Menus 9 REFERENCE PART The following sections describe the ClinProTools menus Section 9 1 and context menus Section 9 2 as well as the MATLAB based menus Section 9 3 9 1 ClinProTools Menus 9 1 1 File Menu The File menu offers the following commands Figure 9 1 Open Model Generation Class Ctrl O Close All Info Loaded Classes Save Class Paths Ctrl S Print Ctrl P Print Preview Print Setup Browse ClinProTools Folder Shift Alt 0 General Settings Exit Figure 9 1 File menu Command Used to Open Model Generation Open
136. e 9 27 Marking the data point that corresponds to the current spectrum by bold display here bold red cross in top left corner 9 1 3 9 Reset View Settings Command The Reset View Settings command resets certain settings of the data plotting views to the default settings Selecting this command opens a confirmation request to reset the view settings Click Yes to reset them click No to keep the current settings The following defaults are restored View Resets to All views default split view partition no grid no auto scaling zooming enabled full display of data zoom reset background of display region white does not apply to Gel View background of axes gray RGB 192 192 192 axis font Arial standard size hidden scales shown again only BMP format to clipboard Spectra View spectra display line no markers line width 1 pixel Gel View displayed color scheme gray scale color intensity quadratic Stack View orientation 30 with basis approx one third of view height colored spectra no whitewash Peak Statistic point width 1 pixel views ClinProTools User Manual Version 2 2 9 29 Reference Part ClinProTools Menus Bruker Daltonik GmbH 9 1 4 Data Preparation Menu The Data Preparation menu offers the following commands Figure 9 28 Data Preparation Settings Spectra Preparation Settings Peak Calculation Load Settings Data Preparation Save Settings Data
137. e Part ClinProTools Menus Bruker Daltonik GmbH Calculate Recognition Capability Check this option if the recognition capability of the generated model should be calcu lated The recognition capability is one measure to describe the performance of a clas sification algorithm It is calculated for a determined model as the relative number of correct classified data points by the classifier for the given model under the constraint that all tested data is previously used for the determination of the model or training of the classifier Calculate Cross Validation Check this option if cross validation should be performed on the generated model using the cross validation procedure selected in Mode Note It is strongly recommended to apply one kind of cross validation to verify that the obtained models give valid results on unseen data Mode Select the mode for calculating cross validation Random Selects a random subset of data points taken over all classes and omits it from the model generation procedure Section 6 2 3 The parameters for this mode are specified under Random Parameters K Fold Divides the set of data points into k equal parts and generates k models where each time a different one of the k parts is omitted Section 6 2 3 The parameter for this mode is specified under K Fold Parameters Leave One Out Leaves exactly one data point out and uses the remaining points for model generation Section 6 2 3 Note In gene
138. e Via the Open Model Generation Class command from the File menu you can select a folder and load all spectra contained as one model generation class This operation has to be repeated until all model generation classes you want to open were loaded e Alternatively a spectra import XML file Appendix A 4 can be opened via the Open Spectra Import XML command This automatically loads all spectra in the refer enced folders as different model generation classes with respect to folder definition Before loading a class the available memory is checked against the memory needed to load the selected spectra provided the Check Memory on Load option in the Gen eral Settings dialog Section 9 1 1 12 is set If the memory is insufficient a warning message will appear which asks you whether to continue Upon opening a class all spectra in the selected referenced folder are loaded and pre pared according to the current spectra preparation settings This includes baseline subtraction and normalization of spectra as well as various additional filtering proc esses The loaded spectra are displayed in the Spectra View and Gel Stack View The first loaded collection is referred to as class 1 in the ClinProTools title bar the second as class 2 etc A running spectra loading can be canceled by clicking or _Lancel_ This cancels the current class loading and also closes and unloads all classes opened before 7 4 ClinProTools User Manual Version
139. e an absolute cc gt correlation level with at least one of the peaks which are already part of this group From the remaining peaks additional groups are created Peaks that are not connected to other peaks form a group on their own Both the peaks within the groups and the groups against each other are sorted according to the sort mode of the statistical settings Thereby one could manually deselect highly correlated peaks from the model building stage to simplify the identification of biomarker candidates ClinProTools User Manual Version 2 2 6 31 Basics Bruker Daltonik GmbH For details please refer to R A Becker J M Chambers and A R Wilks The New S Language Wadsworth amp Brooks Cole 1988 6 4 2 2 Receiver Operating Characteristic ClinProTools calculates a Receiver Operating Characteristic ROC curve for each peak within peak calculation The ROC curve gives a graphical overview about specifi city and sensitivity of a test or within ClinProTools an evaluation of the discrimination quality of a peak The sensitivity represents the true positive fraction TPF and the specificity the true negative fraction TNF Section 6 4 3 6 The fraction of false negatives FNF together with the TPF give a sum of 1 100 and the fraction of the false positives FPF together with the TNF also give a sum of 1 100 Note ROC curves can only be generated for the case of two model generation classes because a true false de
140. e best possible prediction method would yield a graph that was a point in the upper left corner of the ROC space i e 100 sensitivity all TP are found and 100 speci ficity no FP are found A completely random predictor would give a 45 degree diago nal the so called no discrimination line Thus the closer the ROC curve follows the left hand border and then the top border of the ROC space the more accurate it is and the closer the curve comes to the diagonal the less useful is the test at discriminating between two populations A more precise way of characterizing this closeness to the diagonal is to look at the area under the ROC curve AUC The area measures dis crimination which is the ability of the test to correctly classify those with and without 6 32 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Basics disease The closer the area is to 0 5 the less useful is the test and the closer it is to 1 0 the better is the test Lagi FPF DETER 0 646 i FNF a relative frequency 0 021 EE Uaa FPF IRESE 0 152 i a relative frequency 0 276 EZY Al 0 FPF 1 ROC curve Figure 6 6 Graphical explanation of the ROC curve In ClinProTools the ROC curve is generated similar as explained above whereby a peak is considered as the random variable which can be interpreted as a test sepa rating two populations The peak area or the intensity of the peak represents the thresho
141. e end of the branches Node numbers corresponding to the nodes in ClinProtClustering tree xml are displayed at the dendrogram nodes In the additional output ClinProtClustering tree2 xml the XML structure is identical to the tree structure each node is represented by a XML element with two sub node elements The number of sub nodes is available as an attribute at each node If a limited number of classes has been chosen the class number at the end of the branches corresponds to the class number in ClinProtClustering xml The hierarchical clustering uses the MATLAB algorithms For a further documentation of the parameters one can refer to the corresponding MATLAB documentation available at htto Awww mathworks com search for pdist for the Distance Method parameters and for linkage for the Linkage Method parameters For an easy to understand introduction in hierarchical clustering in general see http home dei polimi it matteucc Clustering tutorial_html hierarchical html 6 4 2 5 Pattern Matching for Outlier Detection If peak picking on single spectra is chosen Section 6 1 1 5 2 the overall averaged peak list as well as per class averaged peak lists are calculated If a statistical model is generated these data will be serialized too When spectra are classified against this model a peak list of the spectrum will be generated using the same peak picking parameters used during model generation This peak list will be matched against the
142. e should be aware that the optional filter process needs additional time during the spectra loading step but may be very helpful to improve subse quently processing steps ClinProTools User Manual Version 2 2 6 9 Basics Bruker Daltonik GmbH The following quality filters will be applied to the spectra with the order listed below if enabled Null spectra exclusion filter In seldom cases it happens that due to a preparation artifact or some I O problems a spectrum contains no data or the intensities are extremely low In that case the obtained spectrum cannot be processed in a useful way and should be removed The null spectra exclusion filter identifies such spectra and removes them from the corre sponding class The spectrum is not further processed and cannot be re included with out deselection of the filter and reloading the class Noise spectra exclusion filter The noise spectra exclusion filter aims on identification of noisy spectra Due to an inappropriate preparation or measurement distortions spectra with a high amount of noise may be measured These spectra should be excluded to avoid interfering effects on the further processing This filter checks a given spectrum in the range of 1 Da 4 kDa or if the spectrum does not contain this range the check is done on the whole spectrum It analyses the noise function of the considered spectrum and tries to esti mate the objective amount of noise within the spectrum This esti
143. e the suffix csv the first line must contain M Z Intensity while the following lines must contain the m z intensity pairs separated by a comma M Z Intensity 92 665657 16 709906 92 853842 16 697838 93 042196 17 022505 93 230717 16 194437 93 419405 14 812662 93 608261 13 874493 XML Spectra Import Details of the ClinProtSpectralmport xml import format with path list of spectra to be loaded for statistic calculation and model generation are as follows lt ClinProtSpectraImport Version 0 0 gt lt Class Name Class A gt lt Element Path C A 0 M16 1SLin fid gt Element Path C A 0_M17_1SLin fid gt lt Element Path C A 0_M18 1S5Lin fid gt lt Class gt lt Class Name Class B gt sElement Path C B 0 M19 1Sian fid gt lt Element Path C B 0_M20 1SLin fid gt lt Class gt lt ClinProtSpectraImport gt It is also possible to provide only class paths Name attribute is ignored in this case lt ClinProtSpectraImport Version 0 0 gt lt Class Path C A gt lt Class Path C B gt lt ClinProtSpectraImport gt Mixed formats containing both class and spectra paths are not allowed In both cases an optional RGB attribute can be set changing the displayed class color lt ClinProtSpectraImport Version 0 0 gt lt Class Path C A RGB 255 128 255 gt A 12 ClinProTo
144. eceeeeeeeeeeccneaeeeeeeesetesnanaeeeeees 8 6 8 1125 Model List Rep rter ioe ereua enera reee e a eRe hameateeees 8 7 8 1 1 6 gt Models pOttes e 0 5 kisses eae Gein Sie ese ae aim sadder 8 8 8 1 1 7 Validation Reports i iss0 sek Gigs esate e aeae e ROES r iesean 8 9 vi ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Contents 8 1 1 8 Classification Report 0 2 0 0 ccccccecsceeeceeeeeeeeeceeecaeeeeeeesesensieaeeeeees 8 10 8 11 97 Error Reports ninan i Si eitchs a id ached ee a ee 8 11 8 12 SAVING a Repora erans aaao e a Pies eE aE aa 8 12 1 3 Printingial Repossession a AR tt ede 8 12 8 2 Printing a Graphic of a Data Plotting View 00 cece eee eeee enter ee enteeeeeeenaeeeeetnaeeeeeeaaes 8 12 8 3 Copying a Graphic of a Data Plotting View a PCA Plot or a Dendrogram 8 13 8 4 Exporting the Peak List to XML or CART Format 0 cccceeeeeeeeeeeeeceeeeeeenteeeeeeaes 8 14 9 REFERENCE PART a r aaa aadar ae te a raaraa apaa arana aa ea Aana aea tardaba eels 9 1 9 1 lt ClinPro Tools Menus 2 2 nese latina aan ag eaaa 9 1 9 11 File MOU ievkencdi vestige bee Lael ee es eee 9 1 9 1 1 1 Open Model Generation Classes Command 2 ceeeeeeeeees 9 2 9 1 1 2 Open Spectra Import XML Command ccceceeeeeeeeeeteeeeeeeees 9 3 9 1 1 3 Cancel Commanda ccccccecceceeeeecececeeeeeseeeecaeceeeeeeeeseeenninaeeeeees 9 4 9 1 1 4 Close All Command cccc
145. ed for peak ranking and as weight Difference Average Sorts and weights the peaks by the difference between the maximal and the minimal average peak area of all classes P Value T Test ANOVA Sorts the peaks by the p value Section 6 4 1 6 from t test Section 6 4 1 1 ANOVA test Section 6 4 1 2 P Value Wilcoxon Kruskal Wallis Sorts the peaks by the p value from Wilcoxon test Section 6 4 1 3 Kruskal Wallis test Section 6 4 1 4 OK Opens the Model Name dialog Section 9 1 5 2 5 to specify a name for the model 9 1 5 2 5 Model Name Dialog The Model Name dialog Figure 9 40 is used to specify a name for the new model to be entered in the model list Model Name Enter the Model Name here Figure 9 40 Model Name dialog Entering a model name is optionally when the Force Entering Model Name option in the General Settings dialog Section 9 1 1 12 is not set This dialog is also opened when selecting the Edit Model Name command Section 9 2 9 11 for changing the name of a parameter set in the model list that has still not been calculated ClinProTools User Manual Version 2 2 9 49 Reference Part ClinProTools Menus Bruker Daltonik GmbH 9 1 5 3 Calculate Command The Calculate command is used to calculate a model of the state Added present in the Model List View Model generation uses the peak calculation results of all included peaks of the single non excluded spectra and is based on the settings for the selected
146. ed in a darker color than the corresponding single spectra e g in dark red instead of red The class average spectra are hidden by default The command opens the Display of Averages dialog Figure 9 9 to choose the class es for which the corresponding average spectrum should be shown Display of Averages Display Average Spectra for Cancel Help IV Total Average Figure 9 9 Display of Averages dialog Display Average Spectra for Lists all loaded classes To show class average spectra select the respective class es from this list If a currently shown class average spectrum should be hidden again deselect the respective class in this list Total Average Check this option to display the total average spectrum same as Spectra View gt Total Average Spectrum command from View menu OK Shows the average spectrum spectra for the selected class es in the Spectra View Previously shown class average spectra are hidden if not longer contained in the class selection Shortcut Button fhe 9 1 3 6 5 Spectra View gt Noise Spectrum Command The Noise Spectrum command shows hides the calculated noise spectrum used in average spectrum calculation in the Spectra View The noise spectrum is displayed in orange color and is hidden by default ClinProTools User Manual Version 2 2 9 15 Reference Part ClinProTools Menus Bruker Daltonik GmbH 9 1 3 6 6 Spectra View gt Integration Regions Command The Integr
147. eduction filter In order to speed up calculations and to reduce the memory consumption especially for very large data sets a simple way of data reduction is available The reduction of data is achieved be replacing every set of n consecutive data points by the average of these points The number n is given by the data reduction Factor parameter Typically the value should be chosen between 1 no reduction and 10 10 fold reduce tion The greater the factor n is chosen the more features will be smoothed out As a consequence e g shoulder peaks may no longer be resolved Moreover lower noise estimates are obtained for reduced data Best classification results are expected without data reduction Note The data reduction is applied prior to any other data processing and influences all subsequent results 6 1 3 2 Filters Selecting Spectra ClinProTools allows the processing of a large number of spectra including multiple measurements of the same sample Within the set of spectra individual spectra are of different quality regarding noise chemical artifacts level of intensity etc Therefore and to obtain a faster post processing it is recommended to apply some of the supported quality filters upon the loaded spectra The filters aim on selecting only good spectra and excluding those of lower quality Thereby each filter has its own responsibility e g to pass only spectra with a sufficiently small amount of noise Note In general on
148. eeeeeeeeececaeaeeeeeeesetsenueaeeeeeees 9 30 9 1 4 1 Settings Spectra Preparation Command 0 cceceeeeeeeeeeeeeees 9 30 9 1 4 2 Settings Peak Calculation Commanid cc cccceeeeeeereeeenteeeeees 9 37 9 1 4 3 Load Settings Data Preparation Command l n 9 39 9 1 4 4 Save Settings Data Preparation Command ceeeeeeeeeeeees 9 40 9 1 4 5 Reset Settings Data Preparation Command cceeeeeeeeeeeees 9 40 9 1 4 6 Recalibration Command ccccccecceceeeeeeeeeceaeeeeeeeseseenieaeeeeees 9 40 9 1 4 7 Average Peak List Calculation Command s s s 9 41 9 1 4 8 Peak Calculation Commangd cccccccceeececceceeeeeeteteentaeeeeeees 9 41 Model Generation Menu cccccecceeeeeeeenecceeeeeeeeeeeccaeaeeeeeeesessenieaeeeeees 9 42 9 1 5 1 Settings Peak Selection COMMANG ce eeeeeeeeteeeeetteeeteeneeees 9 43 9 1 5 2 New Model Command cccecceeeccececeeeeeeeeceeceeeeeeeseeeeesieaeeeeees 9 44 9 1 5 2 1 Settings Genetic Algorithm Dialog eee 9 45 9 1 5 2 2 Settings Support Vector Machine Dialog 9 47 9 1 5 2 3 Settings Supervised Neural Network Dialog 9 47 9 1 5 2 4 Settings QuickClassifier Dialog ceeeeeteereeeee 9 48 9 1 5 2 5 Model Name Dialog pec 9 49 viii ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Contents 9 1 5 3 Calculate Command cccccccecsecceceeeeeeeeeeccceaeeeeeeetetensieaeeeeeees 9 5
149. eeeeeeeeeeeneeeeeeeeeeeeeeeneeeeeeaees 5 11 5 1 7 2 Changing the Display Range eccccceeeeeeeeeeeneeeeeeeneeeeeenaeeees 5 12 5 1 7 3 Changing the Stack View s Orientation 5 13 5 1 7 4 Resetting the Data Plotting Views 5 14 5 2 MATLAB Based WiINndOWS 32 tsct22 ecb a areae dees aeaea aaa a aa ite aE iad it 5 14 2 1 PEA WNGOWS n a deh et eee eel Oe eee 5 14 DALI PCA MAN Window reion iner tech bocetined EEEE 5 14 5 2 1 2 Single Scores Plot Loadings Plot Window eeeeeeeeeeee 5 15 52 3 Influence WNdOW eaea ae eaaa eaaa aiaa ea E ATENE 5 16 5251 4 Vatance WINdOW e eeina a ea aaa Eaa ER aE 5 17 5 2 2 Dendrogram Window eanne iera aE E ER A 5 17 6 BASICS ON DATA PREPARATION MODEL GENERATION AND SPECTRA CLASSIFICATION WITH CLINPROTOOLG ceceeeeeeeeeeeeeeeeeeees 6 1 6 1 Data Preparationixs etic ected sedate Reade aera ede 6 1 6 1 1 Standard Data Preparation Workflow cccccccececceeeeeeeeeeeeeeeeeeeeeeeeeeeteees 6 1 6 1 1 1 Baseline Subtraction on Spectra ccccceceeceeccceeeeeeeeeenteeeeeeeees 6 2 6 1 1 2 Normalization of Spectra ecececcecceceeeeeeeeeeeceeceeeeeeesecstceeaeeeeeess 6 3 6 1 1 3 Recalibration of Spectra ceeececeeccececeeeeeeseceneeeeeeeeesetennneaeeeeeess 6 3 6 1 1 4 Average Spectra Calculation cccccceecneeeeeeeieeeeetiieeeeeeieeeeeene 6 4 6 1 1 5 Average Peak List Calculation 0 cccecceeeeeeetteeeeeteteeeeetnieeeeeees 6 4 6 1 1 5 1 Peak Picking
150. eeeeeeeeseeeeeeees 9 76 9 2 9 14 Force Peak N into Model Command u ccccccee cc cceeeeeeeeeeeeeseee scene 9 77 9 2 9 15 Grid Command Command c ccceccceceecesseeeeeessaueeeeeeeeeeeeeeeeeeees 9 77 9 2 9 16 Remove Model Command 0 cccceecceeeceseeeesessueeeeeeeeeeeeeeeee sees 9 77 9 2 9 17 Remove Peak N Command ccccecceeeeccseeeeeccssueeeeeeeeeeeeeeeeeees 9 77 9 2 9 18 ROC Curve for Peak N Command ccccccccecccceeeeeeeeeeeeeeeeeeeeees 9 77 9 2 9 19 Save Model As Command ccccceccccecceeeeeeecseeeeeeeesaeeeeeneeeees 9 78 9 2 9 20 Scaling COMMANG aa T T E OOE 9 78 9 2 9 21 Show Error COMMANG Qu ccccccec cece ee eeceeeeceseseeeseesuueeeeeeaaeeeeeneeseeas 9 79 9 2 9 22 Show Model COMMAN cccccccseececeeeeeeseseeessesaueeeeeeeaseeeeneeseees 9 79 9 2 9 23 Show Spectrum Command cccccceceeeeeeeeeceeeeeeesecseceeeeeeeeeees 9 79 9 2 9 24 Variance for Peak N Command ccccccceceeecccseeeeeeeeeeeeeseeeeeeees 9 79 9 2 9 25 View Spectrum Info Command ceeeeeeeeeceeeeeeeeeeeeeeneeeeeeeees 9 80 9 2 9 26 Whitewash Command ccccccceecccceeeeeee ee eeeseeeeseeuaueeeeeeeeeeeneeeees 9 80 9 2 9 27 Zooming COMMANG ccccceeeceeeeceeeceeeeeeeeeccecaeeeeesetessesiaeeeeeees 9 81 9 3 MATLAB Based Memus a aa aeaa A A AE iaa e aaia SADT 9 82 te PES Sa ORRY e 11 NOLE a E P EET ATETA A EEN 9 82 gA Copy COmMand aan R ATAA 9 82 932 VIEW M U a
151. el generation red bars instead of blue ones mark the peaks incorporated in the model selected in the model list Certain peak statistic data average with standard deviation peak distribution box and whiskers Sections 5 2 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH ClinProTools User Interface 9 1 3 6 7 to 9 1 3 6 9 can be displayed for all or a restricted number of peaks All symbols used are colored like the corresponding class The peak s shown in the Peak Statistics View can be marked with black arrows on the top of the Spectra View Section 9 1 3 6 11 arb u D Data Files ClinProTools ClinProTools Test Data EDTA Run0hiSamplew_E10_1SLin fid 1400 1600 1800 2000 2200 2400 2600 miz Figure 5 2 Spectra View after generating a model A single and the total average spectrum are displayed with marking the picked peaks in blue and the peaks incorporated in the model in red additionally certain peak statistic data is shown The Spectra View allows manual exclusion inclusion of unprocessed spectra and picked peaks manual editing of the average peak list and forcing peaks into a model Excluded spectra are displayed with a darker color than the respective included ones e g in dark red instead of red Excluded peaks are marked by gray integration regions instead of blue ones forced peaks by green ones You can switch the Spectra View to distance mode to show m z differences for two selected peaks Section
152. el generation settings XML file Save the current model generation settings to an XML file with specified name Reset the current model generation settings to their defaults 9 42 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Reference Part ClinProTools Menus 9 1 5 1 Settings Peak Selection Command The Settings Peak Selection command is used to define which peaks should be used in model generation By default all picked peaks are used but you can define that only a restricted number of best peaks with respect to the selected sort mode are taken The peak selection settings are applied to the spectra within the peak calculation work flow however the resulting selection will become effective only in model generation The settings are stored with the cross validation GA SVM SNN and QC settings in the SettingsModelGeneration xml file which is updated on each settings change The command opens the Settings Peak Selection dialog Figure 9 34 Note The peak selection settings may strongly influence the quality of the chosen classification algorithm In many cases a reasonable reduction of peaks improves the classification performed by the algorithms Note Apart from the peak selection settings to use in model generation you can define differing settings for calculating peak statistic Section 9 1 8 5 Settings Peak Selection IV Use All Peaks To Use Sort Mode C Difference Average C P value Wil
153. election Figure 9 55 Correlation List dialog default setting 9 72 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Reference Part ClinProTools Context Menus 9 2 9 6 Display Mode Command Gel View The Gel View s Display Mode popup offers commands to define the color scheme and intensity mode of the view 9 2 9 7 Display Mode Command 2D Peak Distribution ROC Curve Single Peak Variance Views The Peak Statistic View s Display Mode popup offers commands to define the line width of points in the views They can be useful for printing if lines are too thin 9 2 9 8 Display Mode Command Spectra View The Spectra View s Display Mode popup offers commands to define how data points are displayed and connected in the view 9 2 9 9 Display Type Command The Display Type popup offers commands to toggle between Gel and Stack View 9 2 9 10 Distance Command The Distance command switches the Spectra View resp the Gel View to distance mode which allows determining m z differences between two selected points in a spectrum The distance cursor displays in that view where distance measurement was launched Distance measurement behaves similar in both views but the shape of the distance cursor differs the procedure described below concerns the Spectra View When you have finished distance measurement you can deactivate the distance mode by clicking the right mouse button Distance measurement in Spectra View Whe
154. eline The advantage of the Convex Hull baseline is that the baseline is a smooth function Usually the Convex Hull baseline is preferred for a mass range of up to 10 kDa 6 2 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Basics 6 1 1 2 Normalization of Spectra All spectra are normalized to their own TIC total ion count Thereby for each spectrum the TIC is determined as the sum of all intensities of the spectrum Subsequently all intensities of this spectrum are divided by the obtained TIC value After this procedure all intensities are in the range of 0 1 For visualization as e g in the Spectra View the intensities maybe scaled by some additional factor 6 1 1 3 Recalibration of Spectra Usually a mass spectrum is presented as a plot showing intensity over m z mass to charge ratio values However the m z values are not obtained directly from a meas urement Instead these values are computed from the time of flight TOF or other raw data by means of a calibration function Here this function is a quadratic mapping between m z and TOF However systematic time shifts can be observed for individual measurements e g due to the height profile of the preparation This finally leads to peak shifts which can easily be observed in the Gel View For this purpose calibration before measurement is necessary For linear profiling spectra using steel or Anchor Chip targets calibration at one position should be suffi
155. eneration the validity of the classification seems to be better in comparison to other algorithms in many cases How the QC works At first for each peak position the class averages of the peak areas are calculated These averages are stored in the model together with the weights determined from the statistical tests For classification at each peak position the reciprocal difference of the peak area and the class averages are calculated and normalized In the next step over all peak positions from these values weighted averages for the classes are calculated To determine the class member ship these weighted averages are compared Parameterization The parameters for the QC are defined in the Settings QuickClassifier dialog Section 9 1 5 2 4 The QC automatically uses the automatic detection mode to determine the best number of peaks to be integrated in the model Section 6 2 1 5 The Sort Mode defines the peak ranking as well as the weights used for averaging In the case of Difference Average the difference between the maximal and minimal average area of all classes is used as the weight in the case of the P values the logarithm of the p value is used Several models containing up to the first 25 peaks of the ranking are compared internally to determine the optimal peak number Classification result Apart from the calculated class membership the classification result contains a likeli ness for each class It is derived from the weighte
156. enus 0000 0 eccceeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeesenaeeesenaeeeeeenaeees 9 70 9 2 8 Model List View Context Menu ccececeecececeeeeeeeeenaaeeeeeeeeeeeesenneaeeeeeess 9 70 9 2 9 Commands Available from Context Menus Onlly ec ceeeeeceeeeeeeeeeneees 9 70 9 2 9 1 Add Peak Command cccccccecceccceeeeeeeeseccaaeeeeeeesesenniaeeeeeees 9 70 9 2 9 2 Auto Scaling Command ceeceeceeeeeeeeeeenteeeeeeeteeeeteeeaeeeeeeaeees 9 71 9 2 9 3 Background Color COMMANG eecceeeeeeteeeeeeteeeeeeenaeeeeeeenaees 9 71 9 2 9 4 Coordinates Command 00 ccececeece cece ee teeeeeeecaeeeeeeeeeteetieaeeeeees 9 72 9 2 9 5 Correlation List for Peak N Command cccceeeeeeetereeeeeeees 9 72 9 2 9 6 Display Mode Command Gel VieW cccceeeeeeeeenteeeeeeneees 9 73 ClinProTools User Manual Version 2 2 ix Contents Bruker Daltonik GmbH 9 2 9 7 Display Mode Command 2D Peak Distribution ROC Curve Single Peak Variance Views eccceeceeeeeeeeteeeeeeteneeeeetteeeeeeens 9 73 9 2 9 8 Display Mode Command Spectra View 9 73 9 2 9 9 Display Type COMMANG eee eset Gresia i aeii 9 73 9 2 9 10 Distance COMMANGA cc ccc narr ra e a r E aa 9 73 9 2 9 11 Edit Model Name Command ccccccccccccsseeecccseeeeeeeeeeeeeseneeeees 9 75 9 2 9 12 Edit Peak N COMMANG eracarri an nni aana 9 75 9 2 9 13 Exclude Include Peak N Command ccccccccceeec
157. eration Class Open Spectra Import XML Cancel Close All Info Loaded Classes Save Class Paths Print Print Preview Print Setup Peak List Export Browse ClinProTools Folder General Settings Exit Used to Open the selected model generation class Open the selected spectra import XML file and load the referenced spectra Cancel any current loading calculation model generation classification process Close and unload all spectra and models Show paths information about the loaded spectra collections Save the paths of the loaded model generation classes as spectra import XML file Print a graphic of the active data plotting view Preview the graphic of the active data plotting view Set up the printer and printing options Export the peak list to XML or CART format Browse the ClinProTools folder Define general ClinProTools settings Close ClinProTools Button Shortcut Ctrl O a Ctrl I iD Cancel E Ctrl S amp Ctrl P gt Shift Alt O ClinProTools User Manual Version 2 2 Appendix Bruker Daltonik GmbH Menu commands Edit menu Copy Exclude Include Spectrum Bitmap to Clipboard Metafile to Clipboard View menu General Toolbar View Toolbar Status Bar Undo Zoom Redo Zoom Spectra View gt gt Single Spectra gt All Single Spectra gt Total Average Spectrum gt Average Spectra gt Noise Spectrum gt Integration Regions gt Average a
158. es References http gim unmc edu dxtests Default htm 6 44 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Workflows in Detail 7 WORKFLOWS IN DETAIL In the following the ClinProTools workflows will be described in detail here not just the default settings are used like described for the basic workflows in Section 4 4 This section includes details on loading and preparing data for model generation generating and validating classification models classifying spectra calculating peak statistic and correlation analysis and performing PCA and unsupervised clustering 7 1 Spectra Loading and Data Preparation All spectra that are loaded for peak statistic calculation model generation or classifi cation have to be prepared either automatically by applying specified parameters or manually by user action During loading the spectra are prepared by applying various spectra modifying and selecting filters Further data preparation concerning spectra recalibration and averaging average peak list calculations as well as peak calculation and selection is performed when launching the respective ClinProTools workflows 7 1 1 Defining the Data Preparation Settings The data preparation settings specify how spectra preparation and recalibration average spectra calculation average peak list calculation and peak calculation is per formed They are specified in the Settings Spectra Preparation and Settings Peak Calculation dialogs
159. es 7 13 7 2 1 4 Calculating a Model cceeeeeeeeeeeeeeeeeeneeeeeeneaeeeseeaeereeeeaaees 7 13 7 2 1 5 Showing a Single MOdel eeccceeeeeeeeeeeeeeeeeeeeeeeseeeaeeeeeeeaeees 7 13 7 2 1 6 Showing All Models in the Model List ce eeeeeeeeeeeeeeeeeeneees 7 14 Tr Saving a Model lt scicvies ita cities Aaa A dete nines 7 14 7 2 1 8 Removing a Single or All Models from the Model List 7 14 7 2 1 9 Loading a MOdel ec cece i aa a E 7 15 7 2 2 Validating a Model Externally 0 eecceceeeeeeeeeeeeeeeeeeeeeeeeeseeneeeeseeaeeeeeeeaees 7 15 3 Spectra Classification snti Seis eeae cha pest te aa da estates 7 16 7 3 1 Changing the Classification MOde ccccccceeeeeeeeeeeeeeeeeeeneeeteeneereeeaeees 7 16 7 3 2 Selecting a Model for Spectra Classification cc eceeeeeeeeeeeeeeteeeeeeneees 7 16 7 3 3 Selecting the Spectra to be Classified and Running Classification 7 17 7 3 4 Saving the Classification Result 0 0 eee eeeeeeeeeeeeneeeseeeeeeeeeneeeeeeaeees 7 17 7 3 5 Showing the Classification Result ce eccceeeeeeeeeeeeeeeeeseeneeeeeeenaeeeseeaeees 7 18 7 3 6 Closing the Classification ccc ceeeeeeeeeeeeeeeeeeeceeeseneeeeseeeaeeeseeeneeeeeeeaees 7 18 7 4 Peak Statistic and Correlation Analysis Calculation 0 cccceecseeeeeeeteeeeeenteeeeeeaes 7 18 7 4 1 Calculating Peak Statistic eeceeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeaeeeeeeaeees 7 18 7 4 2 Calculating Correlation An
160. es should always be presented along with esti mates of effect and associated confidence intervals In peak statistics Section 8 1 1 2 p values are calculated for each picked peak using the corresponding statistical test The p value can be used for selecting peaks for model generation as well as for sorting and showing peak statistic results 6 4 2 Statistical Methods ClinProTools offers various statistic methods to calculate and visualize statistical properties of the underlying data correlation analysis receiver operating characteristic principal component analysis and unsupervised clustering The methods are described in a less formal way as they can be used for mass spectrometric data within ClinPro Tools 6 4 2 1 Correlation Analysis The correlation analysis is used to analyze stochastic relations between random vari ables upon a given sample set In our context the random variable is given by an indi vidual peak and its properties peak area and the sample set is the given set of spec tra ClinProTools supports calculating correlation matrices Section 8 1 1 3 and per peak correlation lists Section 8 1 1 4 A correlation matrix is obtained by comparing each peak in the list peak to each other peak whereas a correlation list results from comparing a selected peak to each other peak in the list In both cases correlation analysis can be calculated over either all classes or only a selected one Algorithms for correlation an
161. esinaeeeeeees 9 12 9 1 3 3 Status Bar Commanderin airain are araa e araa aa aa aana t 9 12 9 1 3 4 Undo Zoom Command ccceeceeeecceceeeeeeeeeeccecaeeeeeeesesensieaeeeeees 9 12 9 1 3 5 Redo Zoom Command cccceeeeeeeee cece eeeeeeeccacaeeeeeeeeeesesieaeeeeees 9 13 9 1 3 6 Spectra View Popup Command cccceeecceeceeeeeeeeeeeeeeeeeeees 9 13 9 1 3 6 1 Spectra View gt Single Spectra Command 9 14 9 1 3 6 2 Spectra View gt All Single Spectra Commanid 9 14 9 1 3 6 3 Spectra View gt Total Average Spectrum Command 9 14 9 1 3 6 4 Spectra View gt Average Spectra Command 9 15 9 1 3 6 5 Spectra View gt Noise Spectrum Commanid 9 15 9 1 3 6 6 Spectra View gt Integration Regions Command 9 16 9 1 3 6 7 Spectra View gt Average amp StdDev Commana 9 16 9 1 3 6 8 Spectra View gt Peak Distribution Command 9 17 9 1 3 6 9 Spectra View gt Box amp Whiskers Command 9 18 ClinProTools User Manual Version 2 2 vii Contents Bruker Daltonik GmbH 9 1 3 6 10 Spectra View gt Outliers for Box amp Whiskers el nalanta lo RE EEEE E ie ee eal Sie 9 19 9 1 3 6 11 Spectra View gt Peak Markers Commanid 9 20 9 1 3 7 Gel Stack View Popup Command c ecceeeeeeeeeeeeeeeeaeeeeeees 9 20 9 1 3 7 1 Gel Stack View gt Class Names Commanid 9 21 9 1 3 7 2 Gel St
162. est ones are marked as included in model generation shown by blue instead of gray integration regions Depending on the current the statistic settings Section 9 1 8 5 and Spectra View settings Section 9 1 3 6 certain peak statistic data are shown for all or only the selected peaks You can cancel the running process by clicking o Dee ClinProTools User Manual Version 2 2 9 41 Reference Part ClinProTools Menus Bruker Daltonik GmbH 9 1 5 Model Generation Menu The Model Generation menu offers the following commands Figure 9 33 Model Generation Settings Peak Selection New Model Load Model Settings Cross Validation Load Settings Model Generation Save Settings Model Generation Reset Settings Model Generation Figure 9 33 Command Settings Peak Selection New Model Calculate Cancel Load Model Clear All Settings Cross Validation Load Settings Model Generation Save Settings Model Generation Reset Settings Model Generation Model Generation menu Used to Define the peak selection settings Add a new model parameter set to the model list this launches selecting algorithm setting algorithm specific model parameters and specifying model name Start model generation Cancel the current loading calculation generation classification process Load the selected model Clear the Model List View Define the cross validation settings Load the selected mod
163. esults of the PCA 7 20 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Workflows in Detail 7 5 2 Viewing PCA Results The results of a PCA can be viewed in the Scores and Loadings plots the Influence plot and the Variance plot All generated data of a PCA is stored as ClinProtPCA xml file in the ClinProTools folder and can be viewed by launching this file This file will be overwritten when running a new PCA 7 5 2 1 Scores Plots and Loadings Plots The PCA main window displays the results of a PCA in eight different 3D and 2D Scores plots and Loadings plots Figure 5 13 By default the scores and loadings concern the first three PCs PC1 PC2 and PC3 which usually explain most of the variance within in the data set The number of calculated PCs complies with the total number of peaks in the average peak list Scores plots The top row shows four Scores plots with variable axis definitions The top left plot is a 3D plot the following Scores plots are 2D plots which show the three selected PCs in all possible combinations The axes of the Scores plots record arbitrary units Within the Scores plots one point represents one spectrum and each plot contains as many points as non excluded spectra are in the used data set s The points are shown in the same color like the spectra in the Spectra View in the example shown in the figure above two model generation classes were used for PCA The Scores plots display for each s
164. f the state of the selected model has been set to ERROR during model generation 9 2 9 22 Show Model Command The Show Model command creates and shows the Model report Section 8 1 1 6 for the selected model and store the data as ClinProtModel number xmI file Shortcut Button __ Show 9 2 9 23 Show Spectrum Command The Show Spectrum command shows in the Spectra View that spectrum that corresponds to the data point you selected in the Spectra View 2D Peak Distribution View or Single Peak Variance View by right clicking on or close to the data point The command is only available if a data point was right clicked 9 2 9 24 Variance for Peak N Command The Variance for Peak n command displays statistic data average with standard deviation peak distribution and or box and whiskers for the area intensity of the peak for the selected peak This command is available when the Single Peak Variance View is active ClinProTools User Manual Version 2 2 9 79 Reference Part ClinProTools Context Menus Bruker Daltonik GmbH 9 2 9 25 View Spectrum Info Command The View Spectrum Info command is used to show specific information about the selected single spectrum The command opens the Spectrum Information dialog which displays the information stored for the current spectrum Figure 9 64 To view information about another spectrum you can keep the dialog open and just select another spectrum in the Spectra View Spectrum Informa
165. fy own settings suitable for your data The settings are automatically stored in the SettingsModelGeneration xml file which is loaded when ClinProTools is started and is updated on each settings change To keep special settings you can save them in an XML file with a specified name 7 2 1 1 1 Adding a Model Parameter Set to the Model List To calculate a new model you have to add a new model parameter set to the model list This includes selecting the classification algorithm setting the algorithm related model parameters and specifying the model name Entering a model name is optionally but can be forced by checking the Force Entering Model Name option in the General Settings dialog Section 9 1 1 12 The model name can still be edited after the parameter set was entered in the model list via the Edit Model Name command from the Model List View context menu but only as long as model calculation is not started To add a model parameter set 1 From the Model Generation menu select New Model or click New 2 In the Choose Algorithm dialog select the classification algorithm to be used Clicking OK opens the corresponding algorithm specific Settings Algorithm dialog 3 Depending on the chosen algorithm define the model parameters for the e GA in the Settings Genetic Algorithm dialog Section 9 1 5 2 1 7 10 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Workflows in Detail e SVM in the Settings Support Vector
166. g Step Pipetting step Preparation Method Pipetting method 8 1 1 9 Error Report The ClinProt Error report ClinProtError txt Figure 8 10 is created and shown using the Show Error command from the Model List View context menu This command is enabled if the selected model has the state ERROR A message informs you about where and why the error has occurred and what you can do SpectraClassificationObjects Occurence in ModelNTBFilter cpp Zeile 623 Funktion BDal SCO CModelNTBFilter GenerateModel Maximal model size is bigger than peak number Choose a smaller maximal number of best peaks or include more peaks Figure 8 10 Error report Example of an error message ClinProTools User Manual Version 2 2 8 11 Reporting Data Bruker Daltonik GmbH 8 1 2 Saving a Report Each report set up in ClinProTools is automatically saved in an XML file with a conse cutively numbered default name and stored in the ClinProTools folder If desired you can save a shown report via the browser s Save As command either in the ClinPro Tools folder or at another location Note If you like to store the files at another location it is advisable to store a copy of the corresponding style sheet there too 8 1 3 Printing a Report You can print a shown report using the browser s Print command 8 2 Printing a Graphic of a Data Plotting View You can print a graphic of the current content of the selected data plotting view w o prev
167. g a classification model you have to select the spectra collection to be classified and run classification The spectra are prepared according to the data prepa ration parameters stored in the model and then classified based on the respective model generation parameters The classification workflow depends on the active classification mode e In the standard mode all spectra to be classified are loaded in ClinProTools and dis played in the Spectra Gel and Stack views the class color is black The classifyca tion result is automatically shown in the Classification report Section 8 1 1 8 and stored as ClinProtClassification number xml file The 2D Peak Distribution View dis plays corresponding peak data You can save the result with a specified name Section 7 3 4 In the batch mode no spectra are displayed in the ClinProTools GUI After classifica tion is finished the Save Classification dialog opens to save the classification result in an XML file with a specified name The corresponding Classification report can be shown on demand Section 7 3 5 however in the case of big XML files it is not recommendded to create the report Independent of the mode the software holds the classification as long as you do not close the classification Section 7 3 6 To select a spectra collection and run classification 1 From the Classification menu select Classify 2 In the Browse For Folder dialog navigate to the folder that contains t
168. g the Reset View Settings command from the View menu allows a combined resetting of certain settings for the data plotting views Some resets apply to all views e g state of grid and auto scaling background color of axes axis font data formats for copying graphics others to only a certain one e g gray scale in Gel View orientation of Stack View The complete list of affected settings is given with the description of the command 5 2 MATLAB Based Windows Results obtained by the external MATLAB tool integrated in ClinProTools are presented in the MATLAB based PCA windows and Dendrogram window 5 2 1 PCA Windows The PCA windows display data of a PCA Section 6 4 2 3 The windows originate from the external MATLAB software tool integrated in ClinProTools The PCA main window opens automatically after the PCA is completed Single Scores or Loadings plot windows the Influence window and the Variance window can be shown on demand A once opened PCA window stays open as long as you do not close it or the whole ClinProTools session 5 2 1 1 PCA Main Window The PCA main window Figure 5 13 displays the results of a PCA run performed on the loaded spectra data set s Section 7 5 2 1 It contains eight 3D and 2D plots four Scores plots top row and four Loadings plots bottom row showing the data of three selected principle components PC Section 7 5 2 1 The black crosses in the Loadings plots mark the zero axes Multiple PCA main
169. gt 2D Peak Distribution gt ROC Curve and gt Single Peak Variance commands from the View menu The 2D Peak Distribution View is displayed by default Switching to ROC Curve or Single Peak Variance View is possible after peak calculation was performed however the ROC Curve View can be activated only for the case of two loaded classes 5 1 3 1 2D Peak Distribution View The 2D Peak Distribution View Figure 5 5 displays the distribution of two selected peaks in the non excluded spectra of the loaded model generation classes The peak numbers and m z values are indicated on the x and y axes When a classification was performed in standard mode the view additionally displays the respective peak data for the classified spectra Pk 9 1467 Da 0 20 40 60 80 Pk 16 1898 Da Figure 5 5 2D Peak Distribution View displaying the distribution of two peaks in the spectra from 5 model generation classes and a classified collection the ellipses represent the standard deviation of the peak area class average The data is shown on a two dimensional plane By default the first two best separating peaks of the current statistic sort order are displayed Depending on the peak calculation settings the x axis shows the peak area intensity values with respect 5 6 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH ClinProTools User Interface to the most important peak in accordance to the sort criterion e g its p value and t
170. h the view to whitewash mode Background Color Define the background color of the display region of views 9 2 4 2D Peak Distribution View Context Menu The 2D Peak Distribution View context menu offers the following commands Command Used to Coordinates Show Hide the display of cursor coordinates in the status bar Grid Show Hide the grid in the view Scaling Pop up scaling commands for the view Zooming Activate Deactivate the zoom in mode in the view Undo Zoom Same as Undo Zoom command from View menu 9 68 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Reference Part ClinProTools Context Menus Command Used to Redo Zoom Same as Redo Zoom command from View menu Display Mode Pop up display modes for the view Background Color Define the background color of the display region of views Select Peaks Same as Peak Statistics View gt 2D Options gt Select Peaks command from View menu Show Spectrum Show in the Spectra View the spectrum that corresponds to the right clicked data point 9 2 5 ROC Curve View Context Menu The ROC Curve View context menu offers the following commands Command Used to Coordinates Show Hide the display of cursor coordinates in the status bar Grid Show Hide the grid in the view Display Mode Pops up display modes for the view Background Color Define the background color of the display region of views 9 2 6 Single Peak Variance View Context Menu The Single
171. he y axis the peak area intensity values for the second most important peak respectively If the sort criterion is changed the 2D Peak Distribution View may change too because there may now be other peaks that are considered as most important The axis measures are given in arbitrary units which are chosen automatically to fit the plot optimal in the plane You can change the default peak selection Section 9 1 3 8 5 1 All data points belonging to the same model generation class resp to the classified spectra collection are displayed with the same symbol colored according to the class color e g red cross class 1 spectra green diamond class 2 spectra or black dia mond classified spectra If multiple measurements are used and spectra grouping is enabled the peaks of all spectra of a group are averaged before they are processed by the algorithms However the 2D Peak Distribution View does not show the averaged peaks but the peaks from all spectra The ellipses can represent the standard deviation of the class average of the peak areas intensities or the 95 confidence interval which is the standard deviation weighted by the reciprocal number of data points Section 9 1 3 8 5 2 Classified spectra of course are shown without such statistic information 5 1 3 2 ROC Curve View The ROC Curve View Figure 5 6 Sensitivity displays the Receiver Operating Characteristic ROC curve Section 6 4 2 2 for the selected peak generated f
172. he corresponding request Note If the license key for the Support Vector Machine is entered when ClinProTools is started a restart of ClinProTools is necessary to make the Support Vector Machine available 9 1 10 Help Menu The Help menu offers the following commands Figure 9 51 Help Topics F1 About ClinProTools Figure 9 51 Help menu Command Used to Help Topics Launch ClinProTools Help About ClinProTools Show copyright and license information for your ClinProTools installation ClinProTools User Manual Version 2 2 9 65 Reference Part ClinProTools Menus Bruker Daltonik GmbH 9 1 10 1 Help Topics Command The Help Topics command launches ClinProTools Help which is used like other help applications running under Windows Shortcut Key F1 9 1 10 2 About ClinProTools Command The About ClinProTools command shows copyright and license information for your ClinProTools installation Figure 9 52 About Bruker Daltonics ClinProTools Bruker Daltonics ClinProTools my ClinProT ools Version 2 2 Build 73 Copyright C Bruker Daltonik GmbH 2007 This product is licensed to Bruker Daltonik Bruker Daltonik GmbH License key Forum http clinprot bdal de Phone 49 421 2205 432 Fax 49 421 2205 106 E Mail clinprot support bdal de Ww http www bdal de Figure 9 52 About Bruker Daltonics ClinProTools dialog 9 66 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH
173. he defaults Define the spectra preparation and recalibration settings Define the average spectra and peak calculation settings Load the selected data preparation settings XML file ClinProTools User Manual Version 2 2 A 3 Appendix Bruker Daltonik GmbH Menu commands Save Settings Data Preparation Reset Settings Data Preparation Recalibration Average Peak List Calculation Peak Calculation Model Generation menu Settings Peak Selection New Model Calculate Cancel Load Model Clear All Settings Cross Validation Load Settings Model Generation Save Settings Model Generation Reset Settings Model Generation Classification menu Classify External Validation Save Classification Used to Save the current data preparation settings as an XML file with a specified name Reset the current data preparation settings to their defaults Recalibrate spectra and calculates average spectra Calculate average peak list Pick peaks calculate peak areas and peak statistic Define the peak selection settings Add a new model parameter set to the model list Start model generation Cancel any current loading calculation model generation classification process Load the selected model Clear the model list Define the cross validation settings Load the selected model generation settings XML file Save the current model generation settings as an XML file with a spe
174. he peak areas which for the GA SVM and SNN are normalized or the maximal peak intensities can be used Peak selection is performed according to the settings in the Settings Peak Selection dialog Section 9 1 5 1 All picked peaks are taken by default but you can define that only the best peaks according to the chosen sort mode should be selected The peak calculation workflow can be started manually using the Peak Calculation command from the Data Preparation menu The workflow will be run automatically if a workflow that requires peaks being calculated is launched without the peak calculation step has been performed yet 7 8 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Workflows in Detail After peak calculation the integration regions of the peaks selected for model genera tion are marked blue in the Spectra View The 2D Peak Distribution View plots the data of the first two peaks of the peak list by default The results of peak calculation can be viewed by setting up the Peak Statistic report Section 8 1 1 2 The calculated statistic cal data average and standard deviation 1D peak distribution box and whiskers can be shown in the Spectra View on demand Sections 9 1 3 6 7 to 9 1 3 6 9 A running workflow can be canceled by clicking or i This clears all views To continue start peak calculation again or close all spectra and load new classes To calculate and optionally select peaks 1 Specify the pe
175. he same applies to the recognition scores of the spectra of the different classes matched against the corresponding per class average peak list gt ClinProt Model BRUKER Name acl Date Time 2007 04 25T13 38 00 899 02 00 GUILD Sed69a3c 6f77 4e34 8631 f5b 17alaac42 E ClinProTools Version 2 2 build 28 Model Generation Classes Class 1 D Data Files ClinProTools ClinProTools Test Data Spiked Data Normal Class 2 D Data Files ChnProTools ClinProTools Test Data Spiked Data Spiked Cross Validation Percent Leave Out 20 Number of Iterations 10 Overall 89 79 o Class 1 90 Class 2 89 58 Recognition Capability Overall 100 o Class 1 100 Figure 8 7 Model report section 8 8 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Reporting Data 8 1 1 7 Validation Report The Validation report ClinProtValidation xml Figure 8 8 is created and shown after performing a validation using the External Validation command from the Classifica tion menu or by clicking Validate It contains the external validation results The last part of the table contains the confusion matrix This gives an overview how validation data have been classified Each column of the matrix represents the instances in a predicted class while each row represents the instances in an actual class A perfect classification would have entries only on the diagonal which means that all validation data have been classif
176. he scaling of the x axis in the Gel Stack View is changed the x axis in the Spectra View is always adjusted accordingly Depending on whether the mouse or a scaling command is used the adjustment occurs automatically on releasing the mouse button or on the next right click into the Gel Stack View Contrarily the x axis of the Gel Stack View is kept when the x axis in the Spectra View is changed but you can force the Gel Stack View s x axis to follow the x axis of the Spectra View by enabling the Gel Stack View gt Follow Spectra View Mass Range command from the View menu Auto scaling of y axis of Spectra View or Single Peak Variance View The y axis of the Spectra View or the Single Peak Variance View can be set to auto scaling using the Auto Scaling command from the view s context menu When active the axis scaling is automatically adjusted to fully display the most intense peak in the current mass range Spectra View resp the maximum statistic value of the current peak in the loaded classes Single Peak Variance View Zooming To zoom in an area of the Spectra Gel 2D Peak Distribution or Single Peak Variance View activate the Zooming command in the view s context menu and move the mouse cursor in the view to display the zoom cursor G To select the desired area position the zoom cursor at the desired start point and drag it with the left mouse button held down to the desired end point On releasing the mouse button the enclosed area i
177. he spectra collection to be classified and click OK This runs the classification workflow corresponding to the active mode 7 3 4 Saving the Classification Result The classification result for the selected spectra collection can be saved in an XML file with a specified name In the standard mode you have to call up the saving dialog whereas in the batch mode the workflow opens that dialog automatically Saving the result is possible as long as you do not close the classification Section 7 3 6 To save the classification result 1 From the Classification menu select Save Classification to open the Save Classification dialog if it is not shown automatically This dialog opens with the ClinProtClassification folder as the default storage location ClinProTools User Manual Version 2 2 7 17 Workflows in Detail Bruker Daltonik GmbH 2 Enter the file name or select one from the folder list and click Save If you have selected an existing file name answer the confirmation request to overwrite it 7 3 5 Showing the Classification Result The classification result can be shown in the Classification report Section 8 1 1 8 and stored as ClinProtClassification number xmlI file In the standard mode the workflow automatically creates and shows the Classification report you may show the result again if you closed the report In the batch mode the workflow does not set up the Classification report You can create the report on deman
178. hiskers Command The Outliers for Box amp Whiskers command shows hides the outliers for the per class box amp whiskers plots in the Single Peak Variance View and also in the Spectra View when the Spectra View gt Box amp Whiskers command is active This toggles the box amp whiskers plots between the standard box plot command not active and the modified box plot command active Figure 9 22 For description of standard and modified box plot please refer to Section 9 1 3 6 10 ClinProTools User Manual Version 2 2 9 25 Reference Part ClinProTools Menus Bruker Daltonik GmbH arb u Peak 6 1425 38 Da arb u Peak 6 1425 38 Da 5 0 E 45 40 SiG 3 0 mi PAN z 2 4 Class Figure 9 22 Standard box plot left and modified box plot right indicating the outliers diamonds box triangles not belonging to the 95 of values inside the whiskers 9 1 3 8 5 Peak Statistics View gt 2D Options Popup Command Pointing to 2D Options offers the following commands Figure 9 23 EE v 2D Peak Distribution ROC Curve Single Peak Yariance v Outliers For Box amp Whiskers 2D Options i Select Peaks 95 Confidence Interval v Current Spectrum Marker Figure 9 23 2D Options submenu Command Used to Select Peaks Select two peaks to be displayed in the 2D Peak Distribution View 95 Confidence Display the 95 confidence interval or standard deviation in Interval the 2D Peak Distribution View Current Spec
179. ia the Open Import Spectra XML command The command opens the Save Class Paths as Spectra Import XML dialog with the ClinProtSpectralmport folder as the default storage location Enter the file name or select one from the folder list and click Save If you have selected an existing file name answer the confirmation request to overwrite the file Shortcut Keys Ctrl S 9 1 1 7 Print Command The Print command is used to print a graphic of the content of the active data plotting view The command opens the Print dialog to specify the printer and printing options and start printing Note It may be advisable to limit the printer s resolution to 300 dpi For example a resolution of 600 dpi produces four times the number of data as a resolution of 300 dpi does and a resolution of 1200 dpi even produces sixteen times the number Thus printing will take much longer or even may be stopped when using a higher resolution Where in the Print dialog the resolution can be set depends on the respective printer Note Printing a graphic of the Gel or Stack View may take much time Alternatively you can copy the graphic with only the Bitmap to Clipboard command Section 9 1 2 3 being active to the clipboard paste it into e g Microsoft Paint or PowerPoint and then print it from there Shortcuts Button S Keys Ctrl P 9 1 1 8 Print Preview Command The Print Preview command previews the graphic for the active data plotting view as it would be p
180. ic Algorithm dialog Section 9 1 5 2 1 The basic parameters define the Maximal Number of Peaks in Model and the Maxi mal Number of Generations iterations for the algorithm to run When using the default value 50 for the latter most of the time this value will not be reached as the stop criteria will halt calculation when no better peak combination is found for a number of iterations For K nearest neighbors classification Section 6 2 2 the Number of Neighbors can be set to default odd values The Advanced parameters define how the initial number of peak combinations within a population is determined either by an Automatic Detection mode or by specifying the Number of Peak Combinations The Mutation Rate which is the likelihood of a mutation can be set to values ranging from 0 0 no mutation occurs to 1 0 all peak combinations are mutated in each generation The Crossover Rate which is the likeli hood of a crossover between peak combinations can be set to values ranging from 0 0 no crossovers to 1 0 all peak combinations in each generation are used in crossover and are replaced by their children Since the GA employs random numbers for selection crossover and mutation it is possible and quite likely that different values for most of the parameters especially for Crossover Rate and Mutation Rate may yield different solutions To make comparisons between peak combinations possible this randomness can be made the same for all pea
181. ied to their own class The following data is displayed for each class Column Description Class Classes in the model numbers 1 N Name Path and name of the validation spectrum spectra collection assigned to the respective class Correct Classified Percentage of correctly classified part of valid spectra per class Part of Valid Spectra N Number of spectra classified in predicted class 0 Number of unclassified spectra Inv Number of invalid spectra at present only filled by not recali bratable spectra ClinProt Validation BRUKER Date Time 2007 04 26T08 42 36 075 02 00 ChnProTools Version 2 2 build 28 Correct Classified Part of Class Name Valid Spectra 1 210 Inv 1 D Data Files ChnProTools ChnProTools Test Data Spiked 100 3 010 0 Data External Validation Normal gt D Data Files ClinProTools ClinProTools Test Data Spiked 100 0 310 o Data External Validation Spiked Figure 8 8 Validation report ClinProTools User Manual Version 2 2 8 9 Reporting Data Bruker Daltonik GmbH 8 1 1 8 Classification Report The Classification report ClinProtClassification xml Figure 8 9 is created and shown after performing a classification using the Classify command from the Classification menu or by clicking _Classity It contains the classification per spectrum in a table If the single spectra peak picking approach was used additionally the recognition scores of the classified spectra matched agai
182. ier detection 6 36 Loading 7 11 PC menu MATLAB 9 85 Resetting 7 11 PCA 4 Saving 7 11 Calculating 7 20 Model generation standard workflow 4 4 Description 6 34 Model list Explained variance 7 23 Clearing 7 14 Influence plot 7 22 Showing 7 14 Influence window 5 16 Model List command 9 63 Loadings 6 35 Model List report 7 14 8 7 Loadings plot 7 21 Model List View 5 9 Main window 5 14 Model List View context menu 9 70 Performing 7 20 Model name specifying 7 11 Scores 6 35 Model parameter set adding 7 10 Scores plot i 7 21 Model report 7 13 8 8 Single PCA plot window 5 15 Modified box amp whiskers plot 9 19 Variance plot 7 23 Multiple Hypothesis testing 6 41 Variance window 5 17 Multiple measurements 6 7 6 37 6 40 Viewing result 7 21 XML result file 7 22 N PCA command 9 57 PCA main window 5 14 New Model command 9 44 PCA plot Noise spectra exclusion filter 6 10 9 34 Changing PC selection 7 22 Noise spectrum 9 15 Copying 8 13 Noise Spectrum command 9 15 Copying graphic 7 22 Normalization Displaying single plot 7 22 Peak list 6 6 Marking data point 7 22 Spectra 6 3 PCA windows 5 14 Null spectra exclusion filter 6 10 9 34 PCs command MATLAB 9 85 Peak O Adding 7 7 9 70 Calculating 7 8 Open Import Spectra XML command 9 3 Changing integration region 7 7 9 75 Open Model Generation Class Editing 6 11 7 7 9 75 command 9 2 Excluding 7 9 9 76 l 4 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Index Forcing into
183. iewing it Note If the lines in a printout from the Spectra 2D Peak Distribution or ROC Curve View are too thin you can use the Display Mode gt 2 Pixel and 3 Pixel com mands from the view s context menu to display and thus print thicker lines Note It may be advisable to limit the printer s resolution to 300 dpi For example a resolution of 600 dpi produces four times the number of data as a resolution of 300 dpi does and a resolution of 1200 dpi even produces sixteen times the number Thus printing will take much longer or even may be stopped when using a higher resolution Note Printing a graphic of the Gel or Stack View may take much time Alternatively you can copy the graphic to the clipboard with only the Bitmap to Clipboard command being active paste it into e g Microsoft Paint or PowerPoint and then print it from there To print a graphic of a data plotting view 1 Select the view of which you want to print a graphic 2 Depending on whether or not you want preview the graphic proceed as follows e To preview the graphic Select Print Preview from the File menu This sets up the graphic in the preview window as it would be printed If you want to print the graphic now click Print and proceed to step 3 Otherwise close the preview window 8 12 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Reporting Data e To directly print the graphic Select Print from the File menu or click S or press the keys Ct
184. igate to the file you want to load and click Open This overwrites the current data preparation settings with the loaded ones If spectra are currently loaded a message informs you on how to proceed 9 1 4 4 Save Settings Data Preparation Command The Save Settings Data Preparation command is used to save the current spectra preparation and peak calculation settings in an XML file with a specified name The command opens the Save Data Preparation Settings File dialog with the Settings DataPreparation folder as the default storage location Enter a file name or select one from the folder list and click Save If you have selected an existing file name answer the confirmation request to overwrite the file 9 1 4 5 Reset Settings Data Preparation Command The Reset Settings Data Preparation command is used to reset the current spectra preparation and peak calculation settings to their defaults Resetting the data prepara tion settings is always possible however if spectra have already been loaded you might have to close the spectra and load them again or repeat the previously process ing depending on which data preparation settings have been changed The command displays a confirmation request to reset to defaults Click Yes to reset the current settings click No to retain them If spectra are currently loaded a message informs you on how to proceed 9 1 4 6 Recalibration Command The Recalibration command is used to run the recalibration wo
185. iles from the ClinProTools folder The command opens the Settings General dialog Figure 9 4 Show Tables With Choose whether you want to view most of the ClinProt xml files with the browser specified in Browser or with Excel Browser Uses the browser selected in Browser Excel 2002 or Higher Uses Excel Note You need Excel 2002 or newer because the older versions do not support XML with style sheets When Excel starts check the option Open the file 9 6 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Reference Part ClinProTools Menus with the following style sheet applied The Excel security settings extras options securities macro security must be set to low To avoid the BRUKER logo to be displayed in Excel which is wrong positioned due to an error in Excel check Hide all in Objects in the View tab in the Tools Options dialog Settings General Show Tables With Browser ance Excel 2002 or Higher Browser Help Internet Explorer 6 0 or Higher Firefox 1 5 or Higher J Suggest Model Name as File Name Force Entering Model Name 7 Check Memory on Load l Check Memory for PCA Classify in Batch Mode J Show Smoothing Warning l Disable MATLAB Clear Temporary XML Files Reset General Settings Figure 9 4 Settings General dialog default setting Browser If Browser is selected in Show Tables With select the browser you want
186. ility e Leave One Out As the name suggests exactly one data point is left out The remaining points are used for model generation The omitted data point is classified against the model This procedure is repeated for n times where n is the number of data points The obtained classification results are stored for the n models averaged and returned as the prediction capability In general the choice of the cross validation procedure depends on the number of available data points For larger data sets a K Fold or Random approach is recommended If the number of data points is rather small e g less than 30 spectra per class and it is expected that a high variation within each class exists it is more reliable to use the Leave One Out method since in that case more data points remain for the modeling stage 6 2 4 External Validation External validation allows similar to the cross validation measure Section 6 2 3 obtained during the model generation procedure predicting the capability of a calcu lated model External validation requires loading new sets of spectra for each class e g control cancer_stage_1 cancer_stage 2 These validation spectra should not have been used in the model generation step and could come e g from a fresh measurement in accordance to the same clinical protocols of patients The validation spectra are loaded and prepared in the same way as the spectra used in model generation and then are classified ag
187. ilter Section 6 1 3 2 how many of these masses can be found in each spectrum All spectra with a Spectrum Quality Value lt Spectrum Quality Threshold are marked as Not Recalibratable and can automatically be excluded ClinProTools User Manual Version 2 2 6 3 Basics Bruker Daltonik GmbH 6 1 1 4 Average Spectra Calculation From the recalibrated preprocessed individual spectra a total average spectrum is calculated The spectra are weighted with the reciprocal size of the classes to get an equal representation of classes with a very different number of spectra Per class aver age spectra are calculated also In the literature J S Morris K R Coombes J Koomen K A Baggerly R Kobaya shi Feature Extraction and Quantification for Mass Spectrometry in Biomedical Appli cations Using the Mean Spectrum Bioinformatics Advance Access 2005 it has been shown that using the mean spectrum is in most cases favorable against using peak lists obtained from the individual spectra 6 1 1 5 Average Peak List Calculation The ClinProTools data analysis workflows rely on peak information determined for each spectrum An average peak list representing all important peaks is calculated at first It contains the start and end positions of these peaks At those peak locations in all single spectra the area or maximal intensity of the peaks is calculated in the next step These peak lists are used as features to determine statistical
188. in fid p j 8827 266fd1dac723 D Data Files ClinProTools ClinProTools Test Sample 102 6 120 9b6a4810 d321 420a Data EDTA Run O00h Sample O_E12_1SLin fid P b682 d443954d7314 D Data Files ClinProTools ClinProTools Test Excluded Sample 102 25 120 c3a6863d 37c5 4c08 Data EDTA Run 00h Sample 0_E9_15Lin fid anp i a975 0dd84daaldb7 D Data Files ClinProTools ClinProTools Test Sample 100 12 120 f3fb2553 33c2 4f2a Data EDTA Run O00h Sample 0_G10_1SLin fid p 96bb 22bea9bc3444 ab395e0e 5ac6 4a5e 840 c802ba8d1c4e D Data Files ClinProTools ClinProTools Test Data EDTA Run 00h Sample 0_G9_1SLin fid Sample 98 09 120 Class 2 02h Sample Mean Laser Name Intensity Shots pie Bs Name State Figure 8 2 Spectra List report section 8 1 1 2 Peak Statistic Report The Peak Statistic report ClinProtStatistic xml Figure 8 3 is created and shown using the Peak Statistic command from the Reports menu or by clicking a This report shows a table with all peaks picked in peak calculation along with several values The total number of peaks and the used sort mode are shown above the table The following data is displayed for each peak Column Description S Inclusion exclusion state of the peak X used for model generation included not used for model generation excluded Index Peak index Mass m z value DAve Difference between the maximal and the minimal average peak area intensity of all
189. in profiling data using Bruker s Biflex Reflex Omniflex Autoflex or Ultraflex mass spectrometers MS ClinProTools combines intuitive visualization features and multiple mathematical algo rithms to generate pattern recognition models for classification and prediction of e g disease from mass spectrometry based profiling data These easy to use software features allow customers to rapidly generate and validate biomarker patterns from their protein profiling data Key features The ClinProTools software has the following key features gt Import of files acquired with Bruker s mass spectrometers import of ASCII file format possible gt Display of averaged and single spectra with intuitive visualization features such as virtual gel view and stack view gt Data processing parameters for baseline subtraction peak definition recalibration normalization etc gt Statistic analysis of peaks from different spectra gt Supervised classification model generation and validation using different sophisticated mathematical and bioinformatic algorithms gt Pattern matching algorithm supporting outlier detection gt PCA and unsupervised hierarchical clustering gt Highlighting of the biomarker location Allows users to visually inspect individual spectrum to verify their results gt Storage of detailed results for each analysis ClinProTools User Manual Version 2 2 1 1 Preface Bruker Daltonik GmbH 1 2 ClinProTools
190. inProTools User Manual Version 2 2 Bruker Daltonik GmbH Reference Part ClinProTools Menus e In batch mode the spectra to be classified are not displayed in the ClinProTools GUI After classification is finished the Save Classification dialog opens to save the classification result in an XML file with a specified name In both modes the classification result is still held by the software as long as the classification is not closed Shortcut Button Classify 9 1 6 2 External Validation Command The External Validation command is used to validate the selected model externally For external validation Section 6 2 4 you should use spectra of which you know the class membership but which were not used to generate the model The command opens the External Validation dialog Figure 9 43 For each class in the current model you have to select validation spectra The data of the validation spectra is prepared as stored in the model The classification result for the validation spectra is shown in the Validation report Section 8 1 1 7 and stored as ClinProtValidation number xml file For a perfect classi fication the confusion matrix would have entries only on the diagonal which means that all validation data have been classified to their own class In addition Classification reports Section 8 1 1 8 can be shown for each class in the model separately the corresponding data is stored as ClinProtClassification number xmI file
191. inProt measurement software automatically collects multiple measurements in a common folder named by the sample_id In general multiple measurements are used to reduce the risk of measure Mutation The random modification of an Individual during the Genetic Algorithm In ClinPro Tools one peak in the individual is replaced by another randomly selected peak ClinProTools User Manual Version 2 2 A 7 Appendix Bruker Daltonik GmbH Normalization The scales for all the features of the spectra are rescaled to one standard Over fitting Over fitting means that an obtained classification model performs much better on the model generation classes than on the test data In general this is an indicator that some parameters during model generation are too strong adapted to the specifics of the model generation data Population A population means a large collection of Individuals within the Genetic Algorithm This may include hundreds to several thousands of individuals Principle Component Analysis Principle Component Analysis PCA is a broadly used mathematical technique designed to extract display and rank the variance within a data set The overall goal of PCA is to reduce the dimensionality of a data set while simultaneously retaining the information present in the data In ClinProTools the PCA reduces the number of dependent variables contained within the spectra set via replacing groups of variables by a single new variable B
192. information as well as classifica tion models In ClinProTools 2 2 two modes for the generation of the average peak list are avail able e The first one is the total average spectrum peak picking approach for the detection of an average peak list like in ClinProTools 2 1 standard approach Thereby the peak picking is applied on the calculated total average spectrum The identified peak regions by means of start and end positions are subsequently mapped to all single spectra Due to the averaging of the spectra the signal to noise for the peak picking procedure is improved and peaks which may be overlooked on a single spectrum due to noise artefacts can be easier detected on the total average spectrum Even small but reproducible peaks will be detectable which would get lost in the noise of single spectra While this approach is promising and works quite well in case of e g two class approaches with similar class sizes it maybe less appropriate in case of a larger number of classes or in case of classes which are very imbalanced This again is due to the averaging property which may not only reduce noise artefacts but may also delete small and rare peaks e To overcome this effect an alternative method for the detection of an overall average peak list the single spectra peak picking approach is supported in ClinProTools 2 2 Very briefly this approach combines multiple peak lists which are obtained by peak pickings on e
193. ing state before after peak picking and whether the cursor is on a peak or between peaks For the Stack View no cursor coordinates are displayed Spectra View 4303 28 Da Pk 32 Y 3 93 arb u Sp 25 X shows the m z value and Y the intensity value If the cursor is on a peak Pk shows the peak number If the cursor is also on a data point Sp shows the number of the corresponding spectrum Gel View 3244 9 Da Y Sp 19 X shows the m z value and Y the spectrum number 5 10 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH ClinProTools User Interface 2D Peak Distribution View X 34 66 Pk 16 1898Da Y 4 98 Pk 15 1882 Da Sp 22 X and Y show the peak area intensity values of the two peaks selected for display together with the peak s number and m z value If the cursor is on a data point Sp shows the number of the corresponding spectrum ROC Curve View 0 523 1 Specificity Y 0 66Sensitivity X shows the 1 Specificity value and Y the Sensitivity value or vice versa depending on the current definition Single Peak Variance View 3 l Y 1 92 arb u Sp 14 X shows the class number and Y the value of the current statistical data If the cursor is on a data point Sp shows the number of the corresponding spectrum 5 1 7 Altering the ClinProTools Data Plotting Views You can alter the display of the data plotting views Spectra Gel Stack 2D Peak Distribution R
194. ing option is available in ClinProTools The similarity selection filter Section 6 1 3 2 can be used to select a characteristic spec trum from the mm which will end in one spectrum per sample and finally the ordinary processing queue as if there would be only single measurements If the similarity selection filter is not used multiple measurements are averaged before further process ing All remaining processing steps e g model generation statistics calculation etc are done upon these averaged spectra one averaged spectrum per sample It is very important to handle mm in this special way If mm are considered as inde pendent measurements spectra grouping not active the statistics will become very inaccurate because the number of spectra is artificially increased e g by a factor of 4 when 4 mm are available per sample and the underlying statistics e g variance etc are in fact invalid In addition the model generation is effected and the cross validation may be inaccurate 6 4 3 4 Dependent Measurements of Different Samples from the Same Clinical Person Another scenario occurs if dependent measurements of different samples from the same Clinical person are used In fact ClinProTools is currently not designed for this purpose and takes no care about dependent samples in a set of spectra The results should be seen under this strong constraint 6 4 3 5 Multiple Hypothesis Testing Analyzing a Large Number of Peaks at the Same Tim
195. ing the best number of peaks to be integrated in a model The automatic mode option Auto matic Detection 1 25 Peaks automatically determines the best number of peaks to be integrated in the model with restricting the number of peaks 1 to 25 peaks For the manual mode the Number of Peaks to be taken has to be specified When the automatic detection mode is active you do not need to manually determine the peak number by iterating the model generation with different settings for the best number of peaks The algorithm does this internally by an automatic iteration To have reliable processing times the search for the number of best peaks is restricted to maxi mal 25 peaks in a model Therefore the automatic peak detection will always create models with 1 to 25 peaks Due to this restriction it could happen that a manually created model with a larger number of peaks or with all peaks included may give better results than a model obtained by automatic detection As a second point the automatic detection incorporates no cross validation hence the best number of peaks is deter mined on the recognition capability only Therefore the obtained model may show over fitting effects The detection mode s that can be applied depend s on the classification algorithm The SVM supports both the automatic and the manual mode the QC and the SNN in principle use the automatic mode whereas the GA always works in manual mode 6 20 ClinProTools User Manual
196. ion 8 1 1 4 is shown automatically after the calculation is finished and stored as ClinProtCorrelationListInumber xm file The correlation matrix workflow automatically runs the spectra recalibration average peak list calculation and or peak calculation workflows if these have not been per formed when launching correlation matrix calculation In contrast per peak correlation list calculation can be done only after peak calculation was performed otherwise the respective command is disabled To calculate a correlation matrix 1 From the Reports menu select Correlation Matrix If peak calculation has not been performed yet the required workflows are run prior to opening the Correla tion Matrix dialog 2 In the Correlation Matrix dialog define the parameters for correlation matrix calculation and click OK This calculates correlation analysis on all peaks and shows the results in the Correlation Matrix report ClinProTools User Manual Version 2 2 7 19 Workflows in Detail Bruker Daltonik GmbH To calculate a per peak correlation list 1 In the Spectra View right click the peak for which you want to calculate correlation analysis and select Correlation List for Peak n 2 Inthe Correlation List dialog define the parameters for correlation list calculation and click OK This calculates correlation analysis on the selected peak and shows the results in the Correlation List report 7 5 Performing PCA To get more informa
197. ion Validate the selected model externally using test spectra for each class Save Classification Save the current classification result in an XML file with a specified name Show Classification Show the classification result for the currently classified spectra in the Classification report Close Classification Close the current classification and in non batch classification mode unloads the classified spectra too 9 1 6 1 Classify Command The Classify command is used to classify a selected spectra collection with the chosen model All spectra are prepared and processed according to the parameter settings stored in the respective model The classification workflow is run according to the active classification mode standard or batch mode Section 6 3 The command opens the Browse For Folder dialog to select the spectra to be classi fied Navigate to the folder of the respective spectra collection select it and click OK to start classification How the classification workflow proceeds depends on the active classification mode e In standard mode the selected spectra are loaded in ClinProTools and display in the Spectra Gel and Stack views with a black class color After the classification the Classification report Section 8 1 1 8 shows the classification result the report is stored as ClinProtClassificationfnumber xml file The 2D Peak Distribution View displays the corresponding peak data for the classified spectra 9 54 Cl
198. ion command Closing All spectra Classification ClinProTools Colored Spectrum State command Coloring of spectrum states Compass menu Confidence interval Confusion matrix Convex Hull baseline Coordinates command Coordinates in status bar Copy command Copy command MATLAB Copying Data plotting view Dendrogram PCA plot Correlation analysis Correlation list calculation Correlation List command Correlation list parameters Correlation List report Correlation matrix calculation Correlation Matrix command Correlation matrix parameters Correlation Matrix report Cross validation Cross validation modes Cross validation parameters Current Spectrum Marker command 2D Peak Distribution View Gel View Customizing data plotting views FANN NAON DY O11 W gt WDY TAART Kin 9 56 7 4 9 4 7 18 9 65 9 21 5 11 D Data acquisition for clinical proteomics 3 1 Data exchange formats Data plotting view Altering Changing display range Copying Customizing Printing Resetting Data preparation Data Preparation menu Data preparation settings Defining Loading Resetting Saving Data preparation standard workflow Data reduction filter Dendrogram Copying Viewing Dendrogram window A 12 5 11 5 12 8 13 5 11 Dependent Measurements of different samples 6 41 Determination of sensitivity specificity 6 42 Display Mode command Display Type command Distance command Distance measuremen
199. ion folder opened by default Navigate to the file you want to load and click Open This overwrites the current model genera tion settings with the loaded ones 9 1 5 9 Save Settings Model Generation Command The Save Settings Model Generation command is used to save the current model generation settings in an XML file with a specified name The command opens the Save Model Generation Settings File dialog with the SettingsModelGeneration folder as the default storage location Enter a file name or select one from the folder list and click Save If you have selected an existing file name answer the confirmation request to overwrite the file 9 1 5 10 Reset Settings Model Generation Command The Reset Settings Model Generation command is used to reset the current peak selection GA SVM SNN QC and cross validation settings to their defaults The command opens a confirmation request to reset to defaults Click Yes to reset current settings click No to retain them ClinProTools User Manual Version 2 2 9 53 Reference Part ClinProTools Menus Bruker Daltonik GmbH 9 1 6 Classification Menu The Classification menu offers the following commands Figure 9 42 Classification Classify External Validation Save Classification Show Classification Close Classification Figure 9 42 Classification menu Command Used to Classify Classify the spectra in the selected collection with the chosen model External Validat
200. ion if the spectra paths should be shown at the end branches singleton nodes Max Path Length Enter the upper limit of spectra path length if Create Full Tree and Show Paths are checked Number of Classes For Create Full Tree unchecked enter up to how many classes the clustering hierarchy should be calculated Show ClinProtClustering xml Check this option if the ClinProtClustering xm l should be launched in the browser if Create Full Tree not checked Advanced gt gt lt lt Shows Hides the advanced parameters Distance Method Select the metric to be used for distance calculation Euclidian Uses Euclidian metric Minkowski Uses Minkowski metric Cosine Uses Cosine metric Correlation Uses correlation metric Spearman Uses Spearman metric Chebychev Uses Chebychev metric ClinProTools User Manual Version 2 2 9 59 Reference Part ClinProTools Menus Bruker Daltonik GmbH Minkowski Exponent For Minkowski metric chosen enter the Minkowski exponent Linkage Method Select the linkage method to be used for distance calculation Average Uses average linkage Ward Uses ward linkage Clicking OK then runs an unsupervised hierarchical clustering on the non excluded spectra in the data set s If the spectra recalibration average peak list calculation and or peak calculation workflows have not been performed yet the respective workflow s will be automatically run before the unsupervised clustering workflow
201. ired 3 Click OK To reset general settings including file open paths statistic and correlation set tings to defaults 1 From the File menu select General Settings 2 Inthe General Settings dialog click Reset General Settings 3 Confirm the request on resetting to defaults 4 Click OK 4 4 Three Basic Workflows in ClinProTools ClinProTools offers three basic workflows Peak Statistic Calculation Model Generation and Classification To get familiar with the ClinProTools user interface and basic processing features we recommend that you run these basic workflows with the ClinProTools demo data from your installation CD simply using the ClinProTools default settings 4 4 1 Basic Workflow Peak Statistic Calculation The basic workflow Peak Statistic Calculation can be used to quickly calculate peak statistics using ClinProTools default settings This workflow includes spectra recalibra tion and average spectra calculation peak picking and peak calculation as well as peak statistic calculation The statistic results are automatically shown in the Peak Statistic report and stored as ClinProtStatistic number xmI file To run the Peak Statistic Calculation workflow 1 Load one or more model generation classes e g Normal and or Spiked from the ClinProTools Test Data folder on the installation CD using the Open Model Generation Class command from the File menu or One class can be loaded
202. is active by default 9 1 2 4 Metafile to Clipboard Command The Metafile to Clipboard command defines that a metafile graphic of the selected data plotting view should be copied to the clipboard when using the Copy command A metafile graphic is copied with resolution of 8000 6000 pixels Whereas the high resolution of the metafile format offers superior graphics quality some programs e g Microsoft Word can get extremely sluggish due to the amount of data when a Gel View graphic is copied as metafile By selecting Tools gt Options gt View gt Show picture place holders from Word s menu you can avoid the redisplay of the graphics on every move 9 1 3 View Menu The View menu Figure 9 7 offers the following commands v General Toolbar v Views Toolbar v Status Bar Spectra View gt Gel Stack view gt Peak Statistics view gt Reset View Settings Figure 9 7 View menu Command Used to General Toolbar Show Hide the General toolbar View Toolbar Show Hide the View toolbar Status Bar Show Hide the status bar ClinProTools User Manual Version 2 2 9 11 Reference Part ClinProTools Menus Bruker Daltonik GmbH Command Used to Undo Zoom Undo last zooming operation Redo Zoom Redo previously undone zooming operation Spectra View Pop up commands for showing data in the Spectra View Gel Stack View Pop up commands for showing data the Gel Stack View Peak Statistics View Pop up
203. is used to select the peak combinations which are most relevant for separation Support Vector Machine SVM This algorithm is motivated from statistical learning theory and is at first used to determine separation planes between the different data classes Upon the obtained planes a peak ranking can be calculated in a second step Supervised Neural Network SNN This algorithm is a prototype based classifica tion algorithm The SNN tries to identify some characteristic spectra for each class which are named prototypes and could be somehow considered as prototypical samples of that class QuickClassifier QC This algorithm is a univariate sorting algorithm The class averages of the peak areas are stored in the model together with some statistical data like the p values at certain peak positions For classification the peak areas intensities are sorted per peak and a weighted average over all peaks is calculated 6 12 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Basics Support Vector Machine SVM QuickClassifier QC pij intensity A Class 2 c2 p2ic2 B oe Class 1 c1 7 WY Separating plane JZ Pins Peaki p1 Peak2 p2 ma Peak ranking Statistics per Geometric characteristics peak Statistical characteristics Genetic Algorithm GA Supervised Neural Network SNN im 1000 1200 1700 2100 2500 Oo Oo Class 1 ge He Class 2 E GA sear
204. isadvantage that critical values must be calculated for each distribution In the case of ClinProTools the AD test has been adapted to test for normal distributions The AD test is an alternative to the chi square and Kolmogorov Smirnov goodness of fit tests The AD test is defined as e HO The data follow a specified distribution in ClinProTools 2 2 normal distribution e Ha The data do not follow the specified distribution The AD test is applicable if at least eight spectra are available It gives an estimate on the normal distribution assumption In general one will consider a large set of peaks with different distribution properties some peaks maybe normal distributed and some are not From a formal point of view the t test ANOVA test can only be used if the underlying distribution fits the normality assumption Hence at first one should look at the p value for the AD test If it is above e g 0 05 one should consider the t test or ANOVA test otherwise the result from Wilcoxon Kruskal Wallis W KW test has to be evaluated The relations are shown in Table 6 1 Table 6 1 Relation of p value from AD to p value from t test ANOVA or W KW p value p value p value p value t tes ANOVA t test ANOVA W KW W KW lt 0 05 gt 0 05 lt 0 05 gt 0 05 p value AD gt 0 05 interesting uninteresting interesting uninteresting peak peak peak peak p value AD 0 05 not applicable not applicable interesting uninteresting peak
205. ivate the standard mode for classification Section 6 3 Clear Temporary XML Files Removes all temporary ClinProt xml and ClinProt txt files from the ClinProTools folder after you have confirmed the corresponding request Reset General Settings Resets the current general settings including file open paths statistic and correlation settings to defaults after you have confirmed the corresponding request Show Smoothing Warning Check this option if a smoothing warning should appear when you select peak picking on Single Spectra Settings Peak Calculation dialog but smoothing is currently not enabled Settings Spectra Preparation dialog Disable MATLAB Check this option if MATLAB should be disabled Checking this option is not recom mend since without MATLAB it is impossible to run single spectra peak picking PCA and unsupervised clustering 9 8 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Reference Part ClinProTools Menus 9 1 1 13 Exit Command The Exit command is used to close ClinProTools Confirm the confirmation request to quit ClinProTools Shortcut Button application s 9 1 2 Edit Menu The Edit menu Figure 9 5 offers the following commands Eg Copy Ctrl C Copy Ctrl C Exclude Spectrum Include Spectrum v Bitmap to Clipboard v Bitmap to Clipboard Metafile to Clipboard Metafile to Clipboard Figure 9 5 Edit menu when the current spectrum is included left or excluded right Command Used
206. ive class if the model was capable to classify this spectrum and to which class it has been classified For the QuickClassifier we also obtain a likeliness measure which indicates some kind of safety regarding the classifi cation of a spectrum to a specific class 6 3 Spectra Classification For classification of unknown spectra a complete classification model is needed The model contains all information needed to prepare and classify the unknown spectra using the same parameters as were used for model generation Classification depends on the type of the model generating algorithm and the underlying classifier ClinPro Tools supports two classification modes settable in the ClinProTools general settings e Standard mode The standard mode is ClinProTools normal classification mode The spectra to be classified are loaded and displayed in ClinProTools After the clas sification the classification result is automatically shown in the Classification report Section 8 1 1 8 and the 2D Peak Distribution View displays the corresponding peak data for the classified spectra The classification result can be saved in an XML file on demand However since all spectra are kept in the memory the number of spectra that can be classified at a time is limited by the memory size e Batch mode The batch mode is an alternative classification mode overcoming the spectra number limitation of the standard mode Processing a big amount of spectra at a time migh
207. k combinations to be generated by applying Use Varying Random Seed This seeds the random number generator with a different value each time so every model is using different random numbers If this 6 14 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Basics option is not checked the GA uses the same number for initializing the random seed in any peak combination to be calculated This way randomness is disabled and it is easier to study the effect of algorithm parameters Classification result The result of the GA is the peak combination which is proved to separate best between the different classes 6 2 1 2 Support Vector Machine Algorithm The concept of the Support Vector Machines SVM was developed by Vladimir Vapnik V Vapnik Statistical Learning Theory Wiley and Sons New York 1998 and is based on the principle of structural risk minimization SRM The aim of SRM is to mini mize an upper bound on the expected risk over each of the hypothesis classes of the considered problem In our case we have a Classification problem with an expected risk of misclassifications We now are interested on a well modeled classifier with minimal risk For the SVM this is formalized in an optimization problem which can be solved using sophisticated mathematical approaches In the simplest case the SVM helps to determine an optimal hyperplane separating two clouds of data Figure 6 2 Figure 6 2 Determination of the op
208. k of two loaded model generation classes This requires first a decision to be made whether class 1 or class 2 should be treated as positive The decision remains valid as long as the ROC Curve View is not closed but can be changed by selecting the command again and making a new decision The current decision also applies to the ROC Curve for Peak n command from the Spectra View context menu 9 24 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Reference Part ClinProTools Menus The command opens the ROC dialog Figure 9 21 to specify the class to be treated as positive Click ing the corresponding button opens the ROC Curve Treat Gass 1 as Poskiva View and displays the ROC curve for the current peak Treat Class 2 as Positive Shortcut Button foc Figure 9 21 ROC dialog 9 1 3 8 3 Peak Statistics View gt Single Peak Variance Command The Single Peak Variance command switches the Peak Statistics View to Single Peak Variance View to display statistic data box and whiskers peak distribution or average with standard deviation for the current peak The data shown depends on the state of the View menu commands Spectra View gt Box amp Whiskers Peak Distribution and Average amp StdDev Section 5 1 3 3 The Single Peak Variance View can also be launched via the Variance for Peak n command from the Spectra View context menu Shortcut Button 134 9 1 3 8 4 Peak Statistics View gt Outliers for Box amp W
209. l data are learned by heart it is easy to predict the class label for this data if they are represented but for unknown data the model may fail to predict the labeling Selection The fittest Individuals are selected and the less capable ones are abandoned during Genetic Algorithm processing This is done by optimizing a cost function which aims on optimal class separation with high variance between classes Using the cost func tion each peak combination is rated by an expense factor which is used as a measure for the fitness Sensitivity The sensitivity is the percentage of correctly classified positives If your aim is to iden tify diseased people sensitivity is the ability to correctly identify those who have the disease the proportion of people with a disease who have a positive test result This measurement can be derived from the model if a two class scenario is analyzed Specificity The specificity is the percentage of correctly classified negatives If your aim is to iden tify diseased people specificity is the ability to correctly identify those who do not have the disease the proportion of people without disease who have a negative test result This measurement can be derived from the model if a two class scenario is analyzed ClinProTools User Manual Version 2 2 A 9 Appendix Bruker Daltonik GmbH Supervised Neural Network The Supervised Neural Network SNN is a prototype based classification algorithm The SNN
210. l populations e Ha alternative hypothesis The samples come from different populations Notice that the hypothesis makes no assumptions about the distribution of the popula tions These hypotheses are also sometimes written as testing the equality of the cen tral tendency of the populations The test statistic for the Kruskal Wallis test is H This value is compared to a table of critical values for U based on the sample size of each group If H exceeds the critical value for H at some significance level usually 0 05 it means that there is evidence to reject the null hypothesis in favor of the alternative hypothesis For details please refer to W H Kruskal and W A Wallis Use of ranks in one criterion variance analysis Journal of the American Statistical Association 47 260 pp 583 621 1952 6 4 1 5 Anderson Darling Test The Anderson Darling test AD test Stephens 1974 is used to test if a sample of data comes from a population with a specific distribution It is a modification of the Kolmo gorov Smirnov KS test and gives more weight to the tails than does the KS test The 6 28 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Basics KS test is distribution free in the sense that the critical values do not depend on the specific distribution being tested The AD test makes use of the specific distribution in calculating critical values This has the advantage of allowing a more sensitive test and the d
211. lable on your computer 7 If the MATLAB Component Runtime is not available the InstallShield Wizard informs you that installing this application is required prior to installing ClinPro Tools Click OK to start MATLAB Component Runtime installation and follow the MATLAB Component Runtime InstallShield Wizard instructions When you are asked whether to install the MATLAB Component Runtime for yourself or for anyone who uses your computer it is recommended to choose Everyone if several users work on this computer Otherwise there might be the problem that the MATLAB Component Runtime will be available only for the user who installed it Click Finish when you get to the end of the wizard prompts 8 After installation of the MATLAB Component Runtime or if it has already been available installing ClinProTools starts Follow the Bruker Daltonics ClinPro Tools 2 2 InstallShield Wizard instructions to set up ClinProTools on your com puter Click Finish when you get to the end of the wizard prompts 9 If you received a license key for ClinProTools it is recommended to activate the license now Section 0 2 3 Supporting More Than 2 GB RAM More than 2 GB RAM are not supported automatically by Windows XP Professional To enforce the usage the 3GB flag has to be set in the boot ini In the case of 4 GB RAM 3 GB are available for the program while 1 GB is reserved for the operation system To avoid too little memory left for the operation
212. lays all spectra of the loaded classes in a three dimen sional space The x axis records the m z value the y axis the peak intensity in arbitrary units and the z axis the loading order The spectra of the first loaded class are in the foreground those of the last loaded one in the background The default orientation of the plot is 30 but you can quickly change it by dragging all axes at once using the mouse Section 5 1 7 3 arb u 100 50 2000 4000 6000 8000 miz Figure 5 4 Stack View displaying the spectra of five model generation classes red green blue ocher violet and of a collection to be classified black The spectra are colored according to their class membership by default Like in the Spectra View excluded spectra are displayed with a darker color than the corre sponding included spectrum e g red gt dark red Excluded spectra can be hidden Section 9 1 3 7 4 The Stack View can be switched to whitewash mode Section ClinProTools User Manual Version 2 2 5 5 ClinProTools User Interface Bruker Daltonik GmbH 9 2 9 26 resulting in a finer structured plot due to resolving overlying structures but hiding the coloring of the class membership of the spectra 5 1 3 Peak Statistics View The Peak Statistics View consists of three views 2D Peak Distribution View ROC Curve View and Single Peak Variance View You can toggle between the views using the Peak Statistics View
213. ld that is used to reach the separation into the two groups ROC curves for all calculated peaks can be viewed in the ROC Curve View as shown in Figure 6 7 On the x axis the 1 specificity in terms of the false positives is given and on the y axis the sensitivity in terms of the true positives is recorded for this it is assumed that the first loaded class is the diseased one and the second loaded class is the non diseased one Both axes are given in values between 0 and 1 At the bottom of the plot the peak number peak position and AUC value are given If the data is separable by a univariate approach considering only one peak as a test criterion the ROC Curve View may already indicate this peak by a high AUC value close to 1 0 Note ROC curves and their AUC values are only estimations and become more con fident with an increasing number of samples ClinProTools User Manual Version 2 2 6 33 Basics Bruker Daltonik GmbH Sensitivity Peak 5 933 519 Da AUC 0 93 0 0 0 2 0 4 0 6 1 Specificity Figure 6 7 ROC curve for a good separating peak with high AUC value 6 4 2 3 Principal Component Analysis ClinProTools offers a statistical data analysis in terms of principal component analysis PCA The PCA is managed by an external MATLAB software tool which is integrated in ClinProTools PCA is a broadly used mathematical technique designed to extract display and rank the variance within a data set The overall goal of
214. le and quite likely that different values for most of the parameters especially for Crossover Rate and Mutation Rate may yield different solutions To make compari sons between peak combinations possible this randomness can be made the same for all peak combinations to be generated Check this option to seed the random number generator with a different value each time so every model is using different random numbers Uncheck this option if the GA should use the same number for initializing the random seed in any peak combination 9 46 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Reference Part ClinProTools Menus to be calculated This way randomness is disabled and it is easier to study the effect of algorithm parameters OK Opens the Model Name dialog Section 9 1 5 2 5 to specify a name for the model 9 1 5 2 2 Settings Support Vector Machine Dialog The Settings Support Vector Machine dialog Figure 9 37 defines the parameters for the SVM The settings are stored as described for the GA settings Section 9 1 5 2 1 Settings Support Vector Machine Peaks in Model V Automatic Detection 1 25 Peaks Number of Peaks j Cancel Help Figure 9 37 Settings Support Vector Machine dialog default setting In Peaks in Model define how the number of best peaks necessary for model genera tion should be determined Automatic Detection 1 25 Peaks Check this option if automatic peak detection Section 6 2
215. lgorithm Choose Algorithm Genetic Algorithm GA Support Vector Machine SVM Cancel Cc a Supervised Neural Network SNN Help QuickClassifier QC Figure 9 35 Choose Algorithm dialog default setting Select the classification algorithm to be used to generate a new model Genetic Algorithm GA Uses the Genetic Algorithm Support Vector Machine SVM Uses the Support Vector Machine Note For usage of the Support Vector Machine a separate license is needed This option is disabled when the license is not present Supervised Neural Network SNN Uses the Supervised Neural Network QuickClassifier QC Uses the QuickClassifier OK Depending on the selected algorithm opens the corresponding dialog for setting algo rithm specific parameters Settings Genetic Algorithm Section 9 1 5 2 1 Settings Support Vector Machine Section 9 1 5 2 2 Settings Supervised Neural Network Section 9 1 5 2 3 or Settings QuickClassifier Section 9 1 5 2 4 dialog Shortcut Button Newt 9 44 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Reference Part ClinProTools Menus 9 1 5 2 1 Settings Genetic Algorithm Dialog The Settings Genetic Algorithm dialog Figure 9 36 defines the basic and advanced parameters for the GA The settings are stored with the peak selection cross valida tion SVM SNN and QC settings in the SettingsModelGeneration xml file which is updated on each settings change Settings
216. linProTools Section 0 After licensing ClinProTools will be started 6 If the SettingsDataPreparation xml and or SettingsModelGeneration xml file is are not available quit information on starting with default values im ClinProTools ClinProTools User Manual Version 2 2 4 1 Getting Started with ClinProTools Bruker Daltonik GmbH 4 2 ClinProTools File Location All files created by ClinProTools 2 2 will be saved to the ClinProTools folder CABDAL ClinProTools_2_2 Files The style sheets like ClinProtModel xls ClinProtClassification xls etc and the default settings files will be installed in this folder too On installation five subfolders are created in the ClinProTools folder which will be opened by the Load Save dialogs by default ClinProtClassifications ClinProtModels ClinProtModelSpectralmport SettingsDataPreparation and SettingsModelGeneration Note If you have previously worked with ClinProTools 2 0 and or ClinProTools 2 1 the corresponding ClinProTools folder C BDAL ClinProTools_ 2 O Files and or C BDAL ClinProTools_2_1 Files will be kept containing amongst others the style sheets for the ClinProTools 2 0 XML and or ClinProTools 2 1 XML files ClinProTools saves temporary ClinProt xml and ClinProt txt files in the ClinProTools folder For example each time the Spectra List or Peak Statistic command is per formed a new temporary ClinProtSpectra xml resp ClinProtStatistic xml file is gener ated ge
217. lot pops up a context menu offering the following command Command Used to Reset to Original View Restore the plot s original view ClinProTools User Manual Version 2 2 9 83 Reference Part MATLAB Based Menus Bruker Daltonik GmbH 9 3 2 4 Rotate 3D Command The Rotate command switches to the rotation mode that allows rotating the 3D plots To rotate a 3D plot move the mouse cursor into the desired plot so that it changes into the rotation cursor gt Click the plot with the left mouse button and while holding the mouse button pressed rotate the plot in the desired direction Double clicking with the left mouse button restores the plot s original view When the command is active right clicking a 3D plot pops up a context menu offering the following command Command Used to Reset to Original View Restore the plot s original view 9 3 3 Plots Menu The Plots menu of the PCA main window offers the following commands Command Used to Variance Display the Variance plot in the Variance window Influence Display the Influence plot for the chosen PC number in the Influence window PC 3D Display the 3D Scores plot in the PC 3D window PCA Display the left 2D Scores plot in the PC A window PCB Display the middle 2D Scores plot in the PC B window PCC Display the right 2D Scores plot in the PC C window Loadings 3D Display the 3D Loadings plot in the Loadings 3D window Loadings A Display the left 2D Loadings plo
218. lues Note If you change the cross validation settings when models of the state Calculated are present in the Models List these models are automatically reset to the state Added In this case model calculation has to be performed again This ensures that all models in the list are based on the same cross validation settings To set the cross validation parameters 1 From the Model Generation menu select Settings Cross Validation 2 In the Settings Cross Validation dialog specify the parameters as desired and click OK If the model list contains models of the state Calculated this resets the models to the state Added 7 2 1 1 3 Saving Loading and Resetting the Model Generation Settings The model generation settings are automatically stored in the SettingsModelGenera tion xml file which is updated on each settings change To keep the model generation settings you have adapted to special analytical tasks you can save them in an XML file with a specified name This allows loading these settings again Changed settings can also be reset to the defaults ClinProTools User Manual Version 2 2 7 11 Workflows in Detail Bruker Daltonik GmbH To save the current model generation settings 1 From the Model Generation menu select Save Settings Model Generation This opens the Save Model Generation Settings File dialog with the SettingsModel Generation folder as the default storage location 2 Specify the file name
219. lusion are indicated by predefined colors In addition not recalibratable but not excluded spectra are marked in pink Table 9 1 lists the colors used and explains their meaning Colored spectrum states are shown by default Table 9 1 Explanation of coloring of spectra in the Gel View and Spectra List Color Description E light gray Null spectrum excluded by null spectra exclusion filter E yellow Spectrum excluded by exception see comment in Spectra List lilac Spectrum excluded by noise spectra exclusion filter Figure 9 17 C green Spectrum excluded by adduct polymer spectra exclusion filter P turquoise Spectrum excluded by similarity selection filter Figure 9 18 red Not recalibratable excluded by spectra quality filter m pink Not recalibratable not excluded C dark gray Spectrum manually excluded Figure 9 17 ClinProTools User Manual Version 2 2 9 21 Reference Part ClinProTools Menus Bruker Daltonik GmbH 2000 4000 6000 8000 miz Figure 9 17 Coloring of spectra excluded by the noise spectra exclusion filter during spectra loading lilac and by manual exclusion respectively dark gray 2000 4000 6000 8000 miz Figure 9 18 Coloring of spectra excluded by the similarity selection filter turquoise the most characteristic spectrum of each group remains included and thus is not specially colored 9 1 3 7 4 Gel Stack View gt Excluded Spectra Command The Excluded Spectra command shows hides exclu
220. masses which occur very frequently within the entire data set is generated The number of these masses is called the Maximum Quality Value After recalibration it is checked how many of these masses can be found in each spectrum using the Maximum Peak Shift parameter as maximum shift The num ber of found masses is called the Spectrum Quality Value The Spectrum Quality Threshold is computed as the product Spectrum Quality Threshold Maximum Quality Value Match to Calibrant Peaks All spectra with a Spectrum Quality Value lt Spectrum Quality Threshold are marked as Not Recalibratable and excluded by using the Exclude not Recalibratable Spectra option The default value of the Match to Calibrant Peaks parameter is set to 30 Some typical values for the Maximum Quality Value are e studies 1 000 10 000 Da Maximum Quality Value 40 80 e studies 8 000 20 000 Da Maximum Quality Value 10 40 e studies 20 000 100 000 Da Maximum Quality Value 4 10 6 1 4 Manual Peak Editing ClinProTools supports manual editing The average peak list can be edited manually by adding or deleting single peaks or changing the integration regions of peaks Section 7 1 5 2 If the peak number is limited to 0 it is possible to create a list consisting only of manual edited peaks If manual peak editing is performed after the peak calculation has already been run a recalculation of the peaks is required ClinProTools User Manual Version 2 2 6 11
221. mation is compared to a user defined Noise Threshold If the estimated noise is too high the spectrum becomes excluded from the set of spectra Excluded spectra can manually be re included Section 7 1 3 but in general the decision of the filter should be appropriate by a valid noise threshold setting Adduct Polymer spectra exclusion filter The samples to measure may contain chemical artifacts like sodium potassium adsorptions or polymer as the most common types The adduct polymer spectra exclu sion filter identifies and excludes spectra which show mass shifts corresponding to one or more specified artifacts Na Mg K PEG and PPG are searched for by default but one can adapt the adduct polymer settings to the analytical task ClinProTools offers two exclusion modes Strict and Less Strict The strict exclusion mode aims on exclusion of spectra which show characteristic shifts in the autocorrelation spectrum with respect to the adduct polymer parameterization The underlying criterion is strict which means if such a shift exists and it is not due to randomness the spectrum is excluded The less strict mode allows spectra with shifts of lower contribution to remain in the spectra set collection This is determined upon an experimental obtained internal threshold Similarity selection filter With the ClinProt equipment it is possible to measure a sample more than one time This is useful since it may happen that a single measurement will
222. menu Variance ao N w b E a ac w a x 50 W o Oo i gt Figure 5 16 Variance window showing a Variance plot 5 2 2 Dendrogram Window The Dendrogram window Figure 5 17 displays the result of an unsupervised hierarchical clustering Section 6 4 2 4 performed on the loaded spectra It originates from the external MATLAB software tool integrated in ClinProTools The Dendrogram window opens automatically after the unsupervised clustering is completed The display of the dendrogram e g full tree w o spectrum paths depends on the clustering parameters used ClinProTools User Manual Version 2 2 5 17 ClinProTools User Interface Bruker Daltonik GmbH im Dendrogram Edit View Test Data EDTA Run O0h Sample0_ G9 est Data EDTA Run O0h Sample 0_G11 est Data EDTA Run O0h Sample 0_G1 Test Data EDTA Run O0h Sample _ E est Data EDTA Run O2h Sample 0_K10715Lin Test Data EDTA Run 02h Sample _I9_TSLin est Data EDTA Run O2h Sample 0_K1 pe n Test Data EDTA Run 04h Sample 0_M971SLin est RatareRTA Run O4h Sample O_M1071SLin ata EDTA RunW8hisampleJ7G1571SLin est Data EDTA RursOeh Sample J Al 4 1SLin est Data EDTA Run O6h Sample_A13_1SLin Figure 5 17 Dendrogram window showing a full tree dendrogram for the EDTA Run demo data and corresponding spectra paths 5 18 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Basics 6 B
223. models only Best Peaks Peak detection mode filled for SVM SNN and QC models SNN Param SNN specific parameter settings filled for SNN models only QC Param QC specific parameter settings filled for QC models only KNN Param Number of neighbors in k NN classification filled for GA SVM and SNN models Column Description X Val Param Cross validation parameter settings Date Time Date and time of model calculation GUID Globally unique identifier of the model ClinProt Model List BRUKER ClinProTools Version 2 2 build 28 Validation GA Param Best Peaks SNNParam SS ENN X Name Algo XVal X1 XD Gap NEP Gen NPC PCs Rate Rate RS HED BxP Cyd NPIs PT Mode omn Mode Gal Ga BO R a 50 true 0 2 0 5 false 3 random svi levee a A true J random sw sus A true 1000 true random Qc1 OG oe ae oe true oa random lt gt Figure 8 6 Model List report section ClinProTools User Manual Version 2 2 8 7 Reporting Data Bruker Daltonik GmbH 8 1 1 6 Model Report The Model report ClinProtModel xml Figure 8 7 is created and shown for the selected model using the Show Model command from the Model List View context menu or by clicking _ It lists all model generation classes parameters and results of the model If the single spectra peak picking approach was used the recognition scores of all the spectra matched against the overall average peak list are also stored in the model and listed in the report T
224. mp StdDev gt Peak Distribution gt Box amp Whiskers gt Outliers for Box amp Whiskers gt Peak Markers Gel Stack View gt gt Class Names Used to Button Shortcut Copy a bitmap and or a metafile Ctrl C graphic of the selected data plotting view to the clipboard Exclude Include the selected spectrum Activate Deactivate the bitmap format for copying graphics to the clipboard Activate Deactivate the metafile format for copying graphics to the clipboard Show Hide the General toolbar Show Hide the View toolbar Show Hide the status bar Undo the last zoom range change in OQ the selected view Redo the last undone zoom range a change in the selected view Pop up the following commands for the Spectra View Show Hide the single spectra ldu Show Hide the overlaid display of all single spectra Show Hide the total average spectrum 4 Show Hide the class average spectra A Show Hide the noise spectrum Show Hide the integration regions Ji Show Hide the average with standard E deviation Show Hide the 1D peak distribution ox Show Hide the box and whiskers E3 Show Hide the outliers for box amp whiskers plots Show Hide the peak markers Pop up the following commands for the Gel Stack View Show Hide the class names in Gel View A 2 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Appendix Menu commands gt Current Spectrum Marker gt Col
225. mplification The PCA ranks the variables accord ing to their influence on the data set Upon PCA calculation the original coordinates of 6 34 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Basics the diagram are transformed to new coordinates ranked by the variance each coordi nate explains The new axes are called PCs B PC1 describes the largest variance within the data set PC2 describes the second largest variance and is orthogonal to PC1 etc This is indicated by the strength and the orientation of the arrows in diagram B The variance explained by a PC is calculated as sum of the individual variance C variance explained variance Figure 6 8 Simplified representation of the generation of PCs from a data set explanations are given in the text From the PCA results so called scores and loadings can be derived and displayed in various plots Section 7 5 2 1 e Scores The score output represents the original data mapped into the new coordinate system which is defined by the PCs Within the Scores plot outliers from a group or from several groups can be discovered and visualized Outliers are samples which are extreme or do not fit the PCA model Independent from the PC coordinates all Scores plots contain the same sample number as the original data set e Loadings During the calculation of PCs the variables peaks obtain different loadings in dependence on their contribution to the explained variance
226. n activating this mode the distance cursor appears in the center of the Spectra View Figure 9 56 This cursor consists of two vertical lines and a two headed arrow One line is fixed whereas the other is moveable and follows the mouse The m z position of the fixed line and the measured m z difference from fixed to movable line are displayed as X and dX values in the status bar You can switch from fixed to moveable line by clicking the left mouse button To measure a distance position the moveable line on the point of the spectrum where you want to start measurement and click the left mouse button This fixes the moveable line at the selected position and switches the previously fixed line to movable Figure ClinProTools User Manual Version 2 2 9 73 Reference Part ClinProTools Context Menus Bruker Daltonik GmbH 9 57 Now place the second line on the point of the spectrum where you want to end your distance measurement Figure 9 58 While moving the line the displayed m z difference value is continuously updated The m z difference is given as positive or negative value depending on the current direction in which you move the moveable line with respect to the fixed line The absolute difference value between both points is always the same You can change the current sign by a left click Figure 9 59 Illustration of m z distance measurement in the Spectra View 100 50 o4 Nees 1450 1500 1550 mz x 1513 31 m z d
227. n case of need The running ClinProTools application can be shut down externally with the help of the Windows Task Manager It is available via the shortcut Strg Alt Del Mark the ClinPro Tools process in Applications or Processes and click End Task resp End Process If the application crashes the process might still be running The process has to be shut down externally otherwise it is not possible to re launch the application This might also be the case if you find it impossible to start the application In the case of a dead lock processing does not end and cannot be cancelled during data preparation or model generation the application has also to be shut down extern ally ClinProTools User Manual Version 2 2 10 1 Error Treatment Bruker Daltonik GmbH If you try to load a very large amount of data it might happen that the computer runs out of memory In this case processing slows down extremely or comes to a standstill Canceling does now longer work then and the process has to be shut down externally 10 2 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Appendix A APPENDIX A 1 Quick Reference on Menus Commands Tool Buttons and Shortcuts in ClinProTools 2 2 The following table lists the menus available in ClinProTools 2 2 and their commands The corresponding toolbar button and shortcut if available and the meaning are also included Menu commands File menu Open Model Gen
228. n option that is not available with the peak selection parameters To calculate peak statistic with without changing peak statistic settings 1 If you want to change the current peak statistic settings select Settings Statistic from the Reports menu Otherwise processed to step 3 2 In the Settings Statistic dialog uncheck Use Selection Sort Mode From Settings Peak Selection Dialog Then select the desired sort mode If you want to display peak statistic data in the Spectra View only for a restricted number of peaks enter the respective peak number Click Peak Statistic to immediately show the Peak Statistic report or click OK to close the dialog with changing the current settings 3 From the Reports menu select Peak Statistic or click as This shows the corresponding Peak Statistic report 7 4 2 Calculating Correlation Analysis A correlation analysis Section 6 4 2 1 can be calculated either for all peaks resulting in setting up a correlation matrix or for a selected peak which creates a per peak correlation list In both cases the correlation analysis can be calculated over either all classes or only a specified one using one of the two correlation algorithms available The result of a correlation matrix calculation is automatically displayed in the Correla tion Matrix report Section 8 1 1 3 and stored as ClinProtCorrelationMatrix num ber xml file For a per peak correlation list calculation the Correlation List report Sect
229. n reference to a newer XML parser version The tool can be downloaded from the Microsoft web site http www microsoft com downloads details aspx FamilyID 1e6185d7 e4e4 43b1 8056 0e5ecd15a88a amp displaylang en or search for Xmlinst exe on the web site To ensure that Excel parses the XML files with style sheet properly make sure that a dot is used as decimal separator by Excel To enforce this go to the Tools Option dialog in Excel On the International tab at Number handling uncheck Use system separators enter a dot as Decimal separator and a comma as Thousands separa tor If this is not set numbers may be parsed as dates and the like e To support more than 2 GB RAM the 3GB flag has to be set in the boot ini Section 2 3 To install ClinProTools 1 Start your Windows application 2 Insert the installation CD into the CD ROM drive of your computer e g E If the Autostart function is activated the CD browser will start automatically and guide you to start the installation You can proceed to step 7 Otherwise if the Autostart feature is not turned on proceed to step 3 5 cia MSAD Click Run 5 In Open type in the command line E setup exe if E is your CD ROM drive 2 2 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Installing and Licensing ClinProTools 6 Click OK This starts installation and checks if the MATLAB Component Runtime is avai
230. n set is used to determine a model by use of the chosen classifier The test set is than used to evaluate the obtained model and to determine the prediction capability This procedure is repeated multiple times and the absolute prediction capabilities are accumulated and finally normalized to a relative prediction capability obtained by the cross validation procedure The kind of splitting into test and model generation set and how the iteration is performed depends on the specific cross validation method It should be noted that a cross validation for very small sample sizes even if only in one class is not very useful and may give unusual results Thus within ClinProTools the cross validation is calculated only if at least 20 not excluded spectra over all groups are available This must also be kept in mind when working with groups of spectra from multiple measurements Section 6 1 2 here at least 20 groups must be available This is since each group is averaged if the similarity selection filter Section 6 1 3 2 is not on and if the similarity selection filter is on also only one spectrum per group is taken For detailed information on cross validation we refer to M J Kearns Y Mansur A Y Ng and D Ron An experimental and theoretical comparison of model selection methods Machine Learning 27 7 50 1997 Parameterization ClinProTools supports three different kinds of cross validation that can be set in the Settings Cross Validation
231. n the Individual Spectra Calculating peaks in the individual spectra is based on the average peak list picked on the total average spectrum or the overall average peak list picked on the single spectra Either the peak areas or the maximal peak intensities can be used for peak calculation which is defined in the Settings Peak Calculation dialog Area calculation is applied by default as peak areas have a smaller variation between spectra than intensities of single points have The area of a peak is obtained by integrating the intensities over the region of the peak according to the selected Integration Type In Zero Level integration the full intensity values are integrated whereas in End Point Level integration only the area above the cutting edge connecting the endpoints is being measured The two options will yield different areas especially for shoulder peaks Depending on the classification algorithm used peak areas may be normalized for being used in model generation Section 6 1 1 7 6 1 1 7 Normalization of Peak Lists for Model Generation With the Genetic Algorithm Section 6 2 1 1 Support Vector Machine Section 6 2 1 2 and Supervised Neural Network Section 6 2 1 3 algorithm the peak lists are normalized before being used in model generation This is necessary to make different peaks comparable to each other Otherwise small peaks would not be treated as equally important 6 6 ClinProTools User Manual Version 2 2 Bruker Dalt
232. nd class average spectra calculation from all non excluded spectra The recalibration of spectra is based on the corresponding settings in the Settings Spectra Preparation dialog Section 9 1 4 1 The recalibration workflow can be started manually using the Recalibration command from the Data Preparation menu The workflow will be run automatically if a workflow that requires spectra being recalibrated is launched without the recalibration step has been performed yet The spectra that are found by the spectra quality filter Section 6 1 3 2 to be not recalibratable are marked as such and become excluded if the corresponding option is set In the Gel View not recalibratable spectra can be marked with a special color code according to their state Section 9 1 3 7 3 different colors are used for not recalibratable but included and not recalibrated excluded spectra The total average spectrum is calculated and shown in the Spectra View by default Section 9 1 3 6 3 The class average spectra are also calculated and can be shown on demand Section 9 1 3 6 4 The same applies to the calculated noise spectrum Section 9 1 3 6 5 A running workflow can be canceled by clicking or ais This clears all views To continue start recalibration again or close all spectra and load new classes To recalibrate spectra and calculate average spectra 1 Specify the recalibration and average spectrum calculation parameters in the Set ting
233. ndicated in the top right corner The view shows the variance for all peaks even if there are only few peaks selected in the statistic settings Section 9 1 8 5 Only one kind of statistic can be displayed at a time The data shown depends on the state of the View menu commands Spectra View gt Box amp Whiskers Peak Distribu tion and Average amp StdDev If none of the commands is active default setting when switching to the Single Peak Variance View automatically the box and whiskers plots are set up in both the Spectra and the Single Peak Variance View If only one com mand is active the corresponding statistical data is shown in the view If more than one command is active a hierarchical order among the three statistics defines which one is displayed at this box and whiskers takes priority over peak distribution and peak distribution again over average with standard deviation Peak 6 1425 38 Da 1 2 3 4 5 Class 1 2 3 4 5 Class Figure 5 7 Single Peak Variance View displaying the box and whiskers with outliers plots left resp the peak distribution plots right for a peak in the spectra from five classes The plots are drawn on a unique y scale that is set to auto scaling by default like in the Spectra View In horizontal direction the box and whiskers and the average with stan dard deviation are spread over the available space and their sizes are automatically adjusted when the window is resized The size of
234. ng if too many proto types are used Internally at least one prototype for each class is used 6 18 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Basics Classification result The final model consists of the final prototypes and the learned metric which can be interpreted as a weighting of input dimensions The classification takes place using only the prototypes in a nearest neighbor approach The weighting is subject of change with respect to correlated peaks This means that if e g two peaks have similar impor tance only one peak will be ranked high and the other peak may vanish For interpreta tion only high weighting values should be analyzed A low ranking value is no indi cation that the peak is unimportant but a high ranked peak is probably important 6 2 1 4 QuickClassifier Algorithm The QuickClassifier algorithm QC is a univariate sorting algorithm The class aver ages of the peak areas are stored in the model together with some statistical data like the p values Section 6 4 1 6 at certain peak positions For classification the peak areas are sorted per peak and a weighted average over all peaks is calculated The QC algorithm has a good performance The univariate approach makes it easy to trace back classification results The classification allows not only determining the class membership but calculates also a likeliness for each class If there are only few samples available for the model g
235. ng selected ones All files can also be saved via the browsers Save As command Section 8 1 2 If you like to store the files at another location it is advisable to store a copy of the style sheet there too A previously created report can be shown again by double clicking its file name in the Windows Explorer This applies to both the reports automatically saved by the system and the reports manually saved with a specified name Note If the XML file you want to show is stored at another location than in the Clin ProTools folder make sure that a copy of the corresponding style sheet suffix xsl is stored in this folder too Concerning problems with empty XML tables please see the installation notes Section 2 2 To avoid the BRUKER logo to be displayed in Excel which is wrong positioned ClinProTools User Manual Version 2 2 8 1 Reporting Data Bruker Daltonik GmbH due to an error in Excel check Hide all in Objects in the View tab in the Tools Options dialog Details of a report will be shown by hovering with the mouse over the table items Figure 8 1 To fit a page to the window in Internet Explorer the text size can be changed via the Text Size command from the Explorer s View menu or with the mouse wheel while holding down the Ctrl button Index Mass D ve PTTA PWKW PAD Peak State x Used For Model Generation Not used TEF i 35E 1 x 3 1466 82 395 11 5 35e 008 0 000461 5 6e 011 Figure
236. nseseeaeenens 2 1 2T System Requiremetls crotar ean e eee ee ee dee 2 1 2 2 Installing ClinProToOlS crnina E AE AAA Er A AE AEN 2 2 2 3 Supporting More Than 2 GB RAM cccccccccccceeeeeeeeceneaeceeeeeeesecacaeeeeeeeeeeseesnaeeeeess 2 3 2 4 Licens Clin Prol OOlSs eraa EE tide ates 2 3 2 5 Uninstalling ClinProTOols sesideman oenining are aieea ia aiai aai 2 5 3 DATA ACQUISITION FOR CLINICAL PROTEOMICS cc cccsssssssseseseeseseens 3 1 Sal INtORUCTIONS stat id acide edhe eran eh BG PR EEE 3 1 3 2 Sample Preparation srira da aeie nee aaea aa aa aeaa aaa Aaa 3 1 3 3 Data Acquisition with flexControl ccccceccceeeeceeceeceeeeeeeeeceeeaeeeeeeeeeesecenaeeeeeeeeneeees 3 2 4 GETTING STARTED WITH CLINPROTOOLS uu cccccccseseessssenneseneseeaeeeeneeeeaas 4 1 41 Starting ClinProTools c ccccnni An ae nn AM ae Gein ani A Ae 4 1 4 2 ClinProTools File Locations raene cc ccsccsccscccsccsssscsssecsecuceuseceveueeuceuseceveueessecueuseauseuses 4 2 4 3 ClinProTools General Settings oc reiriisiiianieann neart iinr inean karaa KTA ERA EKTRE KAN EEEE RRA 4 2 4 4 Three Basic Workflows in ClinProToolsS cccccccceceseseeeeceeeceseeaseceeeeeeeeeaaeeeseeeeeeees 4 3 4 4 1 Basic Workflow Peak Statistic Calculation cccccessececeeeeeeeeseeeeeeeeeeees 4 3 4 4 2 Basic Workflow Model Generation ccccceseecececeeecesceseeeeeeeseaaeeeseeeeeeees 4 4 4 4 3 Basic Workflow Classification
237. nst the overall average peak list as well as against the per class average peak lists are given Classification reports are also created and shown when the Show Single Classifications option is checked for external validation Section 9 1 6 2 For each class a separate report is set up ClinProt Classification BRUKER Spectra Collection Path D Data Files ClinProTools ClinProTools Test Data Spiked Data To Classify 5 5 Model Name acl Date Time 2007 09 14T11 35 57 437 02 00 ClnProTools Version 2 2 build 65 Index Name Classified Class Class Class2 State Score Scorel Score2 D Data Files ClinProTools ClinProTools 1 Test Data Spiked Data To Classify 5 5 true 1 1 97 0 03 0 202 0 167 0 215 0_L15_1SLin_N fid E D Data Files ClinProTools ClinProTools 2 Test Data Spiked Data To Classify 5 5 true 1 1 99 0 01 0 215 0 186 0 185 O_L17_1SLin_N fid D AData Files ChinProTools ClinProTools 3 Test Data Spiked Data To Classify 5 5 true 1 1 97 0 03 0 197 0 152 0 184 0_L19_1SLin_N fid D Data Files ClinProTools ClnProTools 4 Test Data Spiked Data To Classify 5 5 true 1 1 96 0 04 0 236 0 218 0 195 O_M19_1SLin_N fid D Data Files ClinProTools ClnProTools 5 Test Data Spiked Data To Classify 5 5 true 1 1 99 0 01 0 223 0 194 0 194 O_M20_1SLin_N fid D AData Files ClnProTools ClinProTools 6 Test Data Spiked Data To Classify 5 5 true 2 0 16 1 84 0 196 0 125 0 215 O_N14_15Lin_S fid
238. nt selection after wards Section 7 2 1 2 The Settings Peak Selection dialog Section 9 1 5 1 defines the settings for selecting peaks for model generation By default ClinProTools uses all picked peaks in model generation but you can specify that only selected best peaks with respect to the chosen sort mode should be included You can use the default parameters or specify own settings suitable for your data Alternatively you can load a stored model generation 7 2 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Workflows in Detail settings file or reset the current settings to the default values Section 7 2 1 1 3 Note The peak selection settings may strongly influence the quality of the chosen classification algorithm In many cases a reasonable reduction of peaks improves the classification performed by the algorithms Note If the parameters are changed after running the peak calculation workflow the current peak selection is changed according to the new settings To set the peak selection parameters 1 From the Model Generation menu select Settings Peak Selection 2 In the Settings Peak Selection dialog specify the parameters as desired and click OK If the peak calculation workflow has already been run the current peak selection is changed according to the new parameter settings 7 1 1 4 Saving Loading and Resetting the Data Preparation Settings The data preparation settings are automatically sto
239. nual Version 2 2 6 17 Basics Bruker Daltonik GmbH of type green it is assumed that all of its data points belong to the green class too The region which encloses such a set of data points is depicted by a polygon In that way the position of the prototypes in the data space of two peaks induces a partitioning giving the spider net in the example Later classifications are only done upon these prototypes and hence it is important that they are well located in the data space The advantage of the SNN is that it determines local classifier models there can be different regions of prototypes with the same class label and hence it shows good performance for very multimodal data In addition it can determine a peak ranking It naturally deals with multiple classes The drawback is that it aims on empirical risk minimization ERM which may stick in local minima This is reduced by neighborhood cooperation How the SNN works In an initial step a predefined number of prototypes are spread over the data space by use of the Batch Neural Gas algorithm which gives an optimal distribution of the prototypes over the data space in accordance to the data density properties In a second step the SNN optimizes the positions of the prototypes with respect to the class information supervised minimizing the empirical risk Thereby the used metric of the data space is adapted such that dimensions which are relevant for the class separation
240. odel and X Val Calc If an error occurred during model calculation the state ERROR is displayed the kind of error can be viewed in the Error report Section 8 1 1 9 If the cross validation could not be calculated due to not enough spectra lt 20 Insufficient Spectra Number is given under Cross Validation When a model of the state Calculated or Loaded is selected the corresponding peak selection that was used to generate this model is displayed in the Spectra View When no model is selected the Spectra View shows the current peak selection that will be used in a subsequent model generation process You can deselect all models by click ing in the background of the Model List View New calculate Cancel Load Model List Model Name Algorithm Cross Validation Recognition Capability State Date Time Modell GA 97 78 100 00 Calculated 2007 05 16T 09 47 24 162 02 00 Model2 SVM 81 84 100 00 Calculated 2007 05 16T 09 47 24 663 02 00 Model3 SNN 62 39 100 00 Calculated 2007 05 16T 09 47 26 936 02 00 Model4 87 33 100 00 Calculated 2007 05 16T 09 47 29 049 02 00 Models a a Ns Fees ae Figure 5 8 Model List View with five models one is just under calculation ClinProTools User Manual Version 2 2 5 9 ClinProTools User Interface Bruker Daltonik GmbH 5 1 5 Toolbars ClinProTools offers the General and View toolbars Figure 5 9 with buttons for quick mouse access to certain tool
241. ol To copy a graphic of a data plotting view 1 Select the view of which you want to copy a graphic to the clipboard 2 Depending on which type s of graphic should be used activate the Bitmap to Clipboard and or Metafile to Clipboard command s from the Edit menu 3 From the Edit menu of the ClinProTools window select Copy Or click or press the keys Ctrl C This copies the corresponding graphic s to the clipboard 4 Paste the graphic into the desired application If two graphics are on the clipboard the pasting application will take the appropriate one To copy a graphic of a PCA plot or dendrogram 1 From the Edit menu of the PCA window or Dendrogram window select Copy to copy a graphic to the clipboard 2 Paste the graphic into the desired application ClinProTools User Manual Version 2 2 8 13 Reporting Data Bruker Daltonik GmbH 8 4 Exporting the Peak List to XML or CART Format After running peak calculation specific data of the current peak list can be exported to XML or CART format to use the data in downstream applications For XML export three different XML formats are supported Appendix A 4 The XML2 Files format differs from the XML Files format in that it additionally provides the class and spectra paths as attributes The XML3 Files format is similar to the XML2 Files format but a style sheet reference for ClinProtPeakList xsl is added to facilitate working with peak lists in Excel As an alternativ
242. ols 2 2 Integration regions 9 16 Integration Regions command 9 16 K Kendall s tau algorithm 6 31 K nearest neighbor classification 6 21 Kruskal Wallis test 6 28 L LicenseManager command 9 65 Licensing ClinProTools 2 3 Load Model command 9 50 Load Settings Data Preparation command 9 39 Load Settings Model Generation command 9 53 Loading Data preparation settings 7 3 Model 7 15 Model generation settings 7 11 Spectra 7 4 Loadings PCA 6 35 Loadings plot PCA 7 21 M Manual peak editing 6 11 7 7 Mark Data Points command MATLAB 9 82 Mass range filter 6 8 9 33 MATLAB based menus 9 82 Menu reference A 1 Metafile to Clipboard command 9 11 Model Calculating 7 10 7 13 Classification algorithms 6 12 Cross validating 6 22 Generating 4 4 6 12 7 10 7 13 K nearest neighbor classification 6 21 ClinProTools User Manual Version 2 2 l 3 Index Bruker Daltonik GmbH Loading 7 15 Opening Peak number determination modes 6 20 Class 7 5 Removing from model list 7 14 Model generation class 7 5 Saving 7 14 Spectra import XML file 7 5 Selecting 7 16 Outlier detection 6 36 Showing 7 13 Outliers for Box amp Whiskers command State 5 9 Peak Statistics View 9 25 Validating externally 6 24 7 15 Spectra View 9 19 Model generation 4 4 6 12 7 10 Model generation class P Opening 7 5 Model Generation menu 9 42 Pan command MATLAB 9 83 Model generation settings Part numbers A 14 Defining 7 10 Pattern matching for outl
243. ols User Manual Version 2 2 Bruker Daltonik GmbH Appendix lt Class Path C B RGB 128 128 0 gt lt ClinProtSpectraImport gt CART ASCII Peak List Export Details of the CART ASCII format dat by Salford Systems San Diego CA USA are as follows The first line contains the column header The column headed No contains an ongoing index while Class contains the class number the spectrum belongs to The following columns are all prefixed with A_ for peak area resp I_ for peak intensity The mass of the peak is coded into the column title after the prefix with the decimal dot exchanged by an underscore This is an example of the content of a CART export file four peaks two classes three spectra in class 1 two spectra in class 2 where calculated peaks areas were exported No Class A_1467_95 A_1623_92 A_1911_27 A_2779 25 0 1 190 1123 147 5010 224 8864 296 0563 1 1 198 4078 135 5611 223 3063 290 2748 2 1 195 7452 130 1778 217 7962 299 5370 3 2 207 2164 142 4266 228 4644 289 9513 4 2 183 8811 125 1170 211 8117 282 1209 XML Peak List Export Peak list export to XML supports three XML formats lt ClinProToolsPeakLists gt lt ClinProToolsPeakLists2 gt and lt ClinProToolsPeakLists3 gt Details of the XML Files format are as follows lt ClinProToolsPeakLists gt lt Masses gt List of peak masses lt ClassPaths gt Paths of spectra folder lt SpectraPaths gt lt Cl
244. on called fitness function for a large number of solutions which we call peak combinations It considers many possible peak combinations simultaneously Individual An individual is the entity that is artificially evolved by the Genetic Algorithm An indi vidual consists of a set of peaks which is used for kK NN Classification to determine the selective power of this set Individuals can be mutated or subjected to Crossover The individual that is deemed to be best at differentiating the model generation spectra is returned from the Genetic Algorithm as the classification model K nearest neighbor k NN classification The k nearest neighbor k NN classifier algorithm is used within the Genetic Algo rithm and Support Vector Machine to obtain the final classification It just uses the distances between points in the n dimensional space The peak selection is derived from the current GA peak combination or the final SVM peak ranking solution The idea of k nearest neighbor classifiers is to look at the k nearest neighbors and their spectra class membership Model generation data Model generation data are classes of spectra which have been used for the generation of certain models Multiple measurement A multiple measurement is a measurement of the same sample by applying multiple spotting on the target The obtained spectra e g four spectra of the same sample are in general quite similar and must be considered in a common sense The Cl
245. onik GmbH Basics 6 1 2 Spectra Grouping To increase the measurement quality the ClinProtRobot supports multiple measure ment of the same sample Section 6 4 3 3 Such a set is called spotting and consists of multiple spots Spectra belonging to one spotting must be treated different by the software in comparison to independent spectra to avoid errors in the statistical calcula tion To support spectra grouping the Support Spectra Grouping option in the Settings Spectra Preparation dialog must be enabled The spectra grouping parser see below automatically parses spectra paths and groups the spectra according to their sub folder structure Note This option is suitable only for automatically created spectra by the current Clin ProtRobot with the corresponding software If the option is enabled while using a different folder structure the parser might by coincidence detect non existing groups which will lead to calculation errors In each spectra group only one spectrum should be enabled for further processing This can be done automatically using the Enable Similarity Selection option during spectra load which selects the spectrum most similar to the average of a group Alter natively you can manually exclude spectra Section 7 1 3 leaving one per group If more than one spectrum per group is processed due to dependent samples the statistic values might be misleading e g in the case of p values they might be much too low
246. or Machine dialogs Sections 9 1 5 2 1 and 9 1 5 2 2 determines the k i e how many neighbors have to be used in comparisons Per default k can be set only to the odd values 1 3 5 and 7 which has been found to perform reasonable well on different data sets The odd value ensures that in general a classification is obtained using k NN unclassified may still happen for e g three classes and k 3 where two ClinProTools User Manual Version 2 2 6 21 Basics Bruker Daltonik GmbH neighbors belong to different classes and that the solution is sufficiently stable The case of one neighbor k 1 should be used if the number of samples is very small For a larger number of samples per class k gt 1 is recommended Classification result The result of the k NN classification is a neighborhood matrix in a reduced space of the given peak selection and a classification result for each spectrum which has to be classified 6 2 3 Cross Validation Cross validation is a measure for the reliability of a calculated model and can be used to predict how a model will behave in the future It is a method for evaluating the per formance of a classifier for a given data set and under a given parameterization Differ ent methods for cross validation have been proposed The generic principle behind cross validation methods is to split automatically a given set of data into a model generation set and a test set The model generatio
247. ordinates mode is active and the cursor is positioned in a data plotting view the corresponding x and y data is displayed in the status bar The data shown depends on the focused view the processing state and the cursor position with respect to peak position 9 2 9 5 Correlation List for Peak N Command The Correlation List for Peak n command is used to calculate a correlation analysis Section 6 4 2 1 which compares the selected peak to each other peak in the peak list and to create and show the corresponding Correlation List report Section 8 1 1 4 The per peak correlation can be calculated over either all classes or only a specified one The result is stored as ClinProtCorrelationList number xml file The settings for correla tion list setup are stored in the SettingsGeneral xml file which is updated each time you change the correlation settings or any other settings saved to this file Note Resetting the general settings also resets the current correlation settings The command opens the Correlation List dialog to specify how to set up the correla tion List and to start correlation analysis For descritption of the parameters please refer to the Correlation Matrix command Section 9 1 8 3 Clicking OK calculates the correlation analysis over all or the selected class and shows the results in the Correla tion List report Correlation List Use Kendall s Tau Algorithm Cancel V Calculate Over All Classes __ Cancel Class S
248. ored Spectrum State gt Excluded Spectra gt Group Separators gt Follow Spectra View Mass Range Peak Statistics View gt gt 2D Peak Distribution gt ROC Curve gt Single Peak Variance gt Outliers for Box amp Whiskers gt Options gt Select Peaks gt 95 Confidence Interval gt Current Spectrum Marker Reset View Settings Data Preparation menu Settings Spectra Preparation Settings Peak Calculation Load Settings Data Preparation Used to Button Shortcut Show Hide the current spectrum marker in Gel View Mark Do not mark the spectrum state with modified colors in Gel View Show Hide the excluded spectra in Gel Stack View Show Hide the group separators in Gel View Force Do not force the x axis of Gel Stack View to follow the Spectra View mass range Pop up the following commands Switch to 2D Peak Distribution View and display the 2D peak distribution for two selected peaks Switch to ROC Curve View and display the ROC curve for the selected peak Switch to Single Peak Variance View 13h and display peak statistic for the selected peak Show Hide the outliers for box amp whiskers plots in the Single Peak Variance View O Select two peaks to display in the 2D Peak Distribution View Display the 95 confidence interval or the standard deviation Mark Do not mark the data point that corresponds to the current spectrum Reset certain current view settings to t
249. ort mode and whether sorting in groups was performed Note To get the indices and masses that correspond to a table cell hover with the mouse over the table cell of interest ClinProt Correlation Matrix Correlation Calculated Over All Classes B R U K E R ClinProTools Version 2 2 build 28 Correlation Algorithm standard Sort Mode p value tta Sort in Groups false m 19 22 41 38 16 10 5 6 13 12 37 7 N 45 21 4 2 N s _ W Sa A N W _ ao wm a bb ae ao o ol w 7 Se scot oo a ai EE gt K Figure 8 4 Correlation Matrix sepa section ClinProTools User Manual Version 2 2 8 5 Reporting Data Bruker Daltonik GmbH 8 1 1 4 Correlation List Report The Correlation List report ClinProtCorrelationList xml Figure 8 5 is created and shown using the Correlation List for Peak n command from the Spectra View context menu The correlation analysis Section 6 4 2 1 is based on the settings defined in the Correlation List dialog it can be calculated over either all classes or only a specified one The report lists the correlation coefficients cc that were calculated by comparing the selected peak given by its index and m z value above the list to each other peak in the peak list Whether correlation was calculated over all classes or a specified one is reported above the table as well as the used correlation algorithm standard or Kendall s tau
250. ou have to load two classes at least single classes can be loaded for peak statistic operations including PCA One class can be opened at a time The command opens the Browse For Folder dialog Figure 9 2 Navigate to the folder of the class to be opened select it and click OK This loads and prepares the selected spectra and displays them in the Spectra View and Gel Stack View If the Check Memory on Load option Section 9 1 1 12 is set first the available memory is checked against the memory needed to load the spectra If it is insufficient a warning message appears which asks you whether to continue You have to repeat the loading procedure for each class of interest The first loaded collection is referred to as class 1 in the ClinProTools title bar the second as class 2 etc Shortcuts Button Keys Ctri O 9 2 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Reference Part ClinProTools Menus Browse For Folder Open Model Generation Class 1 O Data Files 2 ClinProTools 2 ClinProTools Test Data amp EDTA Run a H G Sample E G 02h E G 04h H 06h ia nah Folder 00h Figure 9 2 Browse For Folder dialog for opening a model generation class 9 1 1 2 Open Spectra Import XML Command The Open Spectra Import XML command is used to open a ClinProtSpectralm port xml and load the referenced spectra accordingly The spectra import XML format can hold either a path list
251. oups are separated by dashed group separator lines 6 1 3 Additional Filters ClinProTools supports additional spectra modifying and selecting filters which can be applied to modify spectra reduce data and exclude spectra of lower quality from further processing Parameters and usage of these filters can be changed in the Settings Spectra Preparation dialog Excluded spectra can be highlighted in the Gel View according to the filter used for exclusion Section 9 1 3 7 3 Note In general one should be aware that the optional filter process needs additional time during the spectra loading step but may be very helpful to improve subse quent processing steps 6 1 3 1 Filters Modifying Spectra ClinProTools supports various spectra modifying filters which will be applied during spectra loading if activated except the recalibration filter that applies to spectra recali bration The Resolution parameter also applies to peak picking on the average spec trum Resolution parameter The peak detection algorithm needs a hint for the peak width to be able to decide what has to be assumed to be a peak and not just an artifact of a broader peak Since the peak width in mass units depends on the mass range and on the TOF instruments resolution is used as a parameter instead resolution mass peak width Defaults for the resolution are given by choosing the respective mass range from the drop down list in the Settings Spectra Preparation dialog
252. overall and per class peak lists stored in the model The resultant scores are 6 36 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Basics displayed in the Classification report Section 8 1 1 8 The range of the scores is 0 1 with O for no correspondence between spectrum and the spectra used during model generation The higher the value the better is the correspondence This value can be used to support outlier detection Spectra with a low overall score respectively with low scores for all classes have a different peak pattern and are there fore candidates for outliers 6 4 3 Remarks on Statistical Problems with MS Data There are certain statistical problems with MS data which are described in the follow ing sections 6 4 3 1 Common Statistical Pitfalls Generic Remarks Statistical constraints Most statistical tests set some constraints on their applicability Some tests e g expect a normal distribution or the variables needs to be independent These constraints must be kept in mind when using the statistical methods within ClinProTools Some tests are robust if the constraints are not completely valid so it is in general safe to additionally process an ANOVA test if the normal distribution is not perfectly given Other methods may be more sensitive and the results may become very inaccurate and invalid Multiple measurements If during the measurements samples are spotted multiple times and are processed
253. ow Hide the class names in the Gel View Current Spectrum Show Hide the current spectrum marker in the Gel View Marker Colored Spectrum State Mark Do not mark the spectrum state with modified colors in the Gel View 9 20 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Reference Part ClinProTools Menus Command Used to Excluded Spectra Hide Show excluded spectra in the Gel View and Stack View Group Separators Show Hide group separators in Gel View Follow Spectra View Force Do not force the Gel View to follow the mass range Mass Range of the Spectra View 9 1 3 7 1 Gel Stack View gt Class Names Command The Class Names command shows hides the names and color coding of the loaded classes in the Gel View The class names are shown by default 9 1 3 7 2 Gel Stack View gt Current Spectrum Marker Command The Current Spectrum Marker command shows hides the current spectrum marker in the Gel View Two arrow markers gt lt at the left and the right border of the Gel View Figure 5 3 indicate the single spectrum that is selected in the Spectra View The current spectrum marker is shown by default 9 1 3 7 3 Gel Stack View gt Colored Spectrum State Command The Colored Spectrum State command shows hides the coloring of spectra according to their state in the Gel View as well as in the Spectra List report Section 8 1 1 1 Different spectrum states concerning automatic exclusion by specific filters or manual exc
254. pecified name When in standard mode you have to select the command to open the saving dialog whereas in batch mode the saving dialog opens automatically within the workflow after classification is finished Saving the classification result is possible as long as you do not close the current classification The command opens the Save Classification dialog with the ClinProt Classification folder as the default storage location Enter the file name or select one from the folder list and click Save If you have selected an existing file name answer the confirmation request to overwrite the file 9 1 6 4 Show Classification Command The Show Classification command is used to show the classification result for the classified spectra collection in the Classification report Section 8 1 1 8 When in standard mode you can use the command to show the result again if you have already closed the automatically created report When in batch mode you can use the command to create the Classification report as it is not automatically created within the workflow however it is not recommended to display big classifications because the browser used for display might take a long time to process the XML file with style sheet Large XML files with style sheet should better be opened in Excel Showing the classification result is possible as long as you do not close the current classification 9 1 6 5 Close Classification Command The Close Classification command is
255. pectra exclusion filter Section 6 1 3 2 Enable Check this option if the noise spectra exclusion filter should be enabled Detected spectra will be excluded Noise Threshold Enter the noise threshold It can be considered as a required mean signal to noise value of the spectra in the considered range Values 1 can be applied the higher the value the stricter the detection will be In Adduct Polymer Spectra Exclusion define the parameters for the adduct polymer spectra exclusion filter Section 6 1 3 2 Enable Check this option if the adduct polymer spectra exclusion filter should be enabled exclusion mode Select at which adduct polymer level the spectra should be excluded Less Strict Excludes only spectra with adducts polymers of a higher level This mode allows spectra with shifts of lower contribution to remain in the spectra set collection This is determined upon an experimental obtained internal threshold Strict Excludes spectra with adducts polymers of any level This mode aims on exclusion of spectra which show characteristic shifts in the autocorrelation spec trum with respect to the adduct polymer parameterization The underlying criterion is strict which means if such a shift exists and it is not due to randomness the spectrum is excluded Advanced Allows defining advanced adduct polymer spectra exclusion parameters The button opens the Settings Adduct Polymer Spectra Exclusion dialog Figure 9 30 to cus
256. pectrum the corresponding scores of the selected PCs They show how the spectra are distributed in the corresponding sub space deter mined by the selected PCs and visualize the relationship between different spectra e g whether different groups are separated from each other or which spectra may be outliers Loadings plots The bottom row shows four Loadings plots with variable axis definitions As for the Scores plots one 3D plot and three 2D plots are given The Loadings plots are coupled to the Scores plots i e the plots below each other belong together and refer to the same PCs The axes of the Loadings plots record the loadings between 1 and 1 Load1 shows the loadings for PC1 Load2 the loadings for PC2 etc Within the Loadings plots each point represents one peak and each plot contains as many points as non excluded peaks are in the average peak list of the used data set s For better visualization black crosses mark the zero axes The Loadings plots display for each peak the loadings the selected PCs They show how principal components are related to the original peaks Peaks that are far away from the central cloud are responsible for the variance within the data set ClinProTools User Manual Version 2 2 7 21 Workflows in Detail Bruker Daltonik GmbH The following operations may help you to better view single data get more details and document results e Each Scores plot or Loadings plot of the PCA main window can be di
257. ption if the peak areas should be used for peak calculation these are determined based on the selected Integration Type Use Intensities Check this option if the maximum peak intensities based on zero level should be used for peak calculation Integration Type For Use Areas selected choose the integration type for calculating peak areas these two options will yield different areas especially for shoulder peaks End Point Level Integrates only the area above the cutting edge connecting the start and end points of the peak Zero Level Integrates the full intensity values OK Changes the current peak calculation settings Depending on the current processing state the views may become cleared to prevent the loaded spectra from further pro cessing then a message will inform you on how to proceed 9 1 4 3 Load Settings Data Preparation Command The Load Settings Data Preparation command is used to load a stored data prepara tion settings XML file Loading a data preparation settings file is always possible how ever if spectra have already been loaded you might have to close the spectra and load them again or repeat the previous processing depending on which data preparation settings have been changed The command opens the Load Settings Data Prepara ClinProTools User Manual Version 2 2 9 39 Reference Part ClinProTools Menus Bruker Daltonik GmbH tion File dialog with the SettingsDataPreparation folder opened by default Nav
258. pup offers commands for changing the display range of a data plotting view Command Used to Expand manually Change axes scaling in the data plotting views based on values entered in the Manual Scaling dialog see below Times 2 Decrease the y range by 2 Divide by 2 Increase the y range by 2 Offset plus Shift the y range up Offset minus Shift the y range down Expand Expand the x range Contract Contract the x range Move left Move the x range to the left Move right Move the x range to the right Reset Reset x and y range to full display of data Manual Scaling dialog Use the Manual Scaling dialog Figure 9 63 to manually change the scaling of x and or y axis in a view The dialog differs depending on the view whre it was launched For the Gel View you can also change the scale of intensity axis and for the Stack View the scale of the z axis spectrum number axis You can enter new values or reset the current values to full display of data in the respective view 9 78 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Reference Part ClinProTools Context Menus Manual Scaling Scaling Y Scaling Intensity Scaling Start EERIE m z Start 0 500 Low wal 0 000e 000 End 10064 0 m z End 5 500 High val 156 29 Reset Cancel Figure 9 63 Manual Scaling dialog here for the Gel View 9 2 9 21 Show Error Command The Show Error command shows the Error report Section 8 1 1 9 i
259. r Daltonik GmbH The views are displayed in split view and are opened by default The active view is marked with a blue selection bar on top You can resize the views by dragging the horizontal or vertical split borders with the left mouse button held down All four views can be changed at once by dragging the split cross Moving borders may result in hiding a view s a hidden view can be shown again by dragging the corresponding border s accordingly You can alter the data plotting views in various ways Section 5 1 7 On closing ClinProTools certain settings for the views are saved and reloaded on next program start The colors that are used to display the single spectra of different class membership and the corresponding peak statistic data as well as the calculated average spectrum spectra and the noise spectrum are predefined in the system All single spectra of a certain class get the same color Class 1 first loaded class spectra are displayed in red class 2 second loaded class spectra in green class 3 third loaded class spectra in blue etc A maximum of ten different class colors is defined if you load more than ten classes class coloring will continue starting again with red Only in case of using spectra import XML files there is the possibility to define own class colors for displaying the spectra of the referenced classes and corresponding data 5 1 1 Spectra View The Spectra View Figure 5 2 displays the single spec
260. r containing this application and perhaps other Bruker Daltonics applications is created in the Start menu s Programs folder Alternatively you can double click the ClinProTools icon created on your desktop during installation If ClinProTools is started without a valid license being present a message informs you that ClinProTools has not been licensed yet Confirming this message automatically starts the Bruker Daltonics LicenseManager On ClinProTools start up the files SettingsDataPreparation xml and SettingsModelGe neration xml are generated containing the corresponding settings On repeated start these files are loaded by the application if present otherwise a new one with default values is generated In addition a file named SettingsGeneral xml is generated in the same way All files are set up in the ClinProTools folder Section 4 2 Starting ClinProTools also initializes MATLAB except the option to disable MATLAB is checked in the General Settings dialog Section 9 1 1 12 For detailed description of the ClinProTools user interface please refer to Section 5 To start ClinProTools from Windows Start menu cio BSED Click Programs Click Bruker Daltonics Click ClinProTools This starts ClinProTools T Administration o h Bruker Daltonics T Manuals gt T Utilities 3 TL FlexAnalysis k 5 If ClinProTools has not been licensed yet confirm the corresponding message and license C
261. r laser power followed by 30 shots on followed by 30 shots on the the same position with same position with approx approx half of the initial only half of the laser power laser power Parameters Ultraflex_ 1 10 kDa Shots 30 the 15 pre shots 30 the 15 pre shots should not be added to the should not be added to the sum spectra from 15 to sum spectra from 15 to 18 18 different positions on different positions on one one anchor sum 450 550 anchor sum 450 550 shots Spectrometer E lon Source 1 25 kV lon Source 2 23 2 kV Pulsed lon Extraction 120 ns 350 ns Polarity positive Matrix Suppression mode gating Gating strength medium high Suppress up to approx 800 Da ClinProTools User Manual Version 2 2 3 3 Data Acquisition for Clinical Proteomics Bruker Daltonik GmbH Parameters Ultraflex_ 1 10 kDa Mass Range low 900 10 500 Da Detector Gain 1600 1800 V Sample Rate 1 00 Electronic Gain regular 100 mV Parameters Ultraflex _ 1 10 kDa Real time smooth high wanena Be careful changes are not saved in the method Laser Frequency 25 Hz 25 Hz Laser Attenuator e g 60 30 e g 60 30 3 4 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Getting Started with ClinProTools 4 GETTING STARTED WITH CLINPROTOOLS 4 1 Starting ClinProTools You can start ClinProTools from Windows Start menu When ClinProTools is installed a Bruker Daltonics folde
262. r of one sample has a unique relationship with a particular member of the other sample e g the same people measured before and after an intervention or IQ test scores of a husband and wife For the classification problems considered with ClinProTools the analyses must be applied on a set of independent individuals where intervention and no intervention are 6 26 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Basics not mixed Therefore we can ignore this variant and consider only independent samples For the kind of multiple measurements of one sample the different spectra are averaged to reduce the measurement variance after a possible pre selection by use of some filter criteria If the calculated t value is greater than the threshold chosen for statistical significance alpha conventionally equal to 0 05 then the null hypothesis that the two groups do not differ is rejected in favor of the alternative hypothesis which typically states that the groups do differ 6 4 1 2 ANOVA Test In statistics analysis of variance ANOVA is a collection of statistical models and their associated procedures which compare means by splitting the overall observed variance into different parts The initial techniques of the analysis of variance were pioneered by the statistician and geneticist Ronald Fisher in the 1920s and 1930s and are sometimes known as Fisher s ANOVA or Fisher s analysis of variance In ClinProTools we
263. r than the included ones As an alternative you can use the Spectra View gt All Single Spectra command to display all single spectra simultane ously and overlaid The separately displayed single spectra are shown by default Shortcut Button 9 1 3 6 2 Spectra View gt All Single Spectra Command The All Single Spectra command shows hides single spectra of the loaded classes in the Spectra View with displaying all single spectra simultaneously and overlaid All single spectra within a class are displayed in the same predefined class color excluded spectra are shown with a darker color than the included ones As an alternative you can use the Spectra View gt Single Spectra command to display only one single spectrum at a time 9 1 3 6 3 Spectra View gt Total Average Spectrum Command The Total Average Spectrum command shows hides the total average spectrum in the Spectra View The total average spectrum Section 6 1 1 4 is calculated from all non excluded spectra within the spectra recalibration process It is displayed in gray color and is shown by default Shortcut Button aha 9 14 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Reference Part ClinProTools Menus 9 1 3 6 4 Spectra View gt Average Spectra Command The Average Spectra command is used to show hide class average spectra in the Spectra View Class average spectra are calculated within the spectra recalibration workflow They are display
264. ral the choice of the cross validation mode depends on the number of available data points For larger data sets a K Fold or Random approach is recommended If the number of data points is rather small e g less than 30 spectra per class and it is expected that a high variation within each class exists it is more reliable to use the Leave One Out method since in that case more data points remain for the modeling stage In Random Parameters define the Random cross validation mode if set Note It is safe to keep these parameters with defaults Percent to Leave Out Enter the percentage of data points to leave out per iteration Number of Iterations Enter the number of iterations to perform In K Fold Parameters define the K Fold cross validation mode if set K Divide in K Parts Enter the number of parts to divide the set of data points 9 52 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Reference Part ClinProTools Menus OK Changes the current cross validation settings If the model list contains models of the state Calculated these models will be reset to the state Added 9 1 5 8 Load Settings Model Generation Command The Load Settings Model Generation command is used to load a stored model generation settings XML file which includes the peak selection GA SVM SNN QC and cross validation settings The command opens the Load Settings Model Genera tion File dialog with the SettingsModelGenerat
265. red in the SettingsDataPrepara tion xml file which is updated on each settings change To keep the data preparation settings you have adapted to special analytical tasks you can save them in an XML file with a specified name This allows loading the settings again Changed settings can be reset to the defaults Loading a data preparation settings file or resetting the current settings to their defaults is always possible however if spectra have already been loaded you might have to close the spectra and load them again or repeat the previ ously processing depending on which data preparation settings have been changed To save the current data preparation settings 1 From the Data Preparation menu select Save Settings Data Preparation This opens the Save Data Preparation Settings File dialog with the SettingsDataPre paration folder as the default storage location 2 Specify the file name and target folder and click Save 3 If you have selected an existing file name answer the confirmation request to overwrite the file To load a data preparation settings file 1 From the Data Preparation menu select Load Settings Data Preparation This opens the Load Settings Data Preparation File dialog with the SettingsDataPre paration folder opened by default 2 Navigate to the file you want to load Double click it or select it and click Open This overwrites the current data preparation settings with the loaded ones 3 If spectra are curren
266. rinted The graphic is set up in the ClinProTools preview window You can click the preview s Print button to print the graphic now otherwise close the preview ClinProTools User Manual Version 2 2 9 5 Reference Part ClinProTools Menus Bruker Daltonik GmbH 9 1 1 9 Print Setup Command The Print Setup command is used to set up the printer and printing options The command opens the Print Setup dialog 9 1 1 10 Peak List Export Command The Peak List Export command is used to export specific peak list data to the CART or XML format For the XML export three different XML formats are available Appen dix A 4 The command opens the Peak List Export dialog to select the export format XML Files XML2 Files XML3 Files or CART Files and specify the file name and target folder Clicking Save exports the peak list data in the selected format 9 1 1 11 Browse ClinProTools Folder Command The Browse ClinProTools Folder command browses the ClinProTools folder C BDAL ClinProTools_2_2 Files Shortcuts Button i Keys Shift Alt O 9 1 1 12 General Settings Command The General Settings command is used to define general non algorithm settings for ClinProTools The general settings are saved to the SettingsGeneral xml file which also stores file open paths statistic and correlation settings The command also allows resetting the settings saved in the SettingsGeneral xml file to the defaults as well as removing all temporary XML f
267. rkflow This includes recalibration of the spectra of the loaded classes and calculation of the total average spectrum and the class average spectra from all not excluded spectra Recalibration is performed based on the recalibration masses already picked during spectra loading and the current recalibration settings Section 9 1 4 1 The command runs spectra recalibration and average spectra calculation The spectra quality filter Section 6 1 3 2 marks spectra which are not recalibratable These become excluded if the corresponding option is set The calculated total average spec trum is shown in the Spectra View by default The additionally calculated class average spectra can be shown on demand Section 9 1 3 6 4 You can cancel the running pro Cancel cess by clicking or 9 40 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Reference Part ClinProTools Menus 9 1 4 7 Average Peak List Calculation Command The Average Peak List Calculation command is used to run the average peak list calculation workflow This automatically picks peaks on either the total average spec trum or the single spectra based on the current peak picking settings Section 9 1 4 2 and determines the integration regions The calculated average peak list can be edited manually Section 7 1 5 2 The command runs the peak picking and average peak list calculation If the recali bration workflow has not been performed yet when selecting this comm
268. rl P 3 Inthe Print dialog select the printer and print options and click OK 8 3 Copying a Graphic of a Data Plotting View a PCA Plot or a Dendrogram You can copy a graphic of the focused data plotting view PCA plot or dendrogram to the clipboard in order to paste the graphic into an appropriate application ClinProTools copies graphics of data plotting views as a bitmap with a resolution of 800 600 pixels by default Alternatively ClinProTools can copy graphics as a metafile with a resolution of 8000 6000 pixels You can define whether a bitmap and or a metafile are is copied via the Bitmap to Clipboard and Metafile to Clipboard com mands from the Edit menu If both types are activated the program that pastes the clipboard s contents into its document determines which of these formats it uses Microsoft Paint uses bitmap by default Microsoft Word Excel and PowerPoint prefer metafile Whereas the high resolution of the metafile format offers superior graphics quality some programs e g Microsoft Word can get extremely sluggish due to the amount of data when a Gel View is copied as metafile By selecting Tools gt Options gt View gt Show picture place holders from Microsoft Word s menu you can avoid the redisplay of the graphics on every move Graphics of PCA plots entire PCA main window single Sores plot single Loadings plot Influence plot and dendrograms are copied as metafiles by the MATLAB to
269. roTools User Manual Version 2 2
270. roTools installation ClinProTools User Manual Version 2 2 A 5 Appendix Bruker Daltonik GmbH A 2 Glossary Class A class is a set of spectra originating from samples e g of the same disease state The model generation classes have to be sorted e g according to the state of disease and are used to generate a model which will be applied to explain the class membership ClinProTools loads all spectra in a folder and its subfolders recursively as one class If the ClinProtRobot is used and multiple spotting takes place it is important to switch on the Support Spectra Grouping option Classification Classification means the determination of the class membership of a given spectrum Classification model A classification model is the result of generating a model It contains data preparation characteristics as well as classifier characteristics It can be used for classification of spectra of unknown status It may be saved as XML file to be reloaded later Classifier algorithm A Classifier means an algorithm used in generating Classification models ClinPro Tools offers four classifier algorithms the Genetic Algorithm the Support Vector Machine the Supervised Neural Network and the QuickClassifier Correlation analysis The correlation analysis is used to analyze stochastic relations between random variables upon a given sample set In our context the random variable is given by an individual peak and its properties peak area
271. rom all included spectra of the loaded two model generation classes The x axis records the 1 specificity in terms of the false positives and the y axis the sensi tivity in terms of the true positives both axes are given in values Peak 2 673 576 Da AUC 0 885 between 0 and 1 To set up a ROC curve a decision has to be made 0 0 0 2 0 4 06 1 Specificity whether class 1 or class 2 should be Figure 5 6 ROC Curve View displaying the treated as positive Section ROC curve for a peak and corre 9 1 3 8 2 sponding data The number and m z value of the peak the ROC curve is displayed for is indicated at the bottom of the plot followed by the ROC curve s AUC value Section 6 4 2 2 If the information is not shown or only partially displayed broadening the view will help You can use the view s scroll bar to browse through the different ROC curves over the present set of peaks ClinProTools User Manual Version 2 2 5 7 ClinProTools User Interface Bruker Daltonik GmbH 5 1 3 3 Single Peak Variance View The Single Peak Variance View Figure 5 7 can display three kinds of statistical data for a selected single peak The box and whiskers with without outliers Section 9 1 3 6 9 peak distribution Section 9 1 3 6 8 or average with standard deviation plots Section 9 1 3 6 7 calculated from the area intensity values of the selected peak in the loaded spectra are shown separately for each class The selected peak is i
272. s Note In the validation workflow each spectrum of the collection will be used there is no selection done by the noise spectra exclusion and adduct polymer spectra exclusion or similarity selection filters Section 6 1 3 2 if there is something detected the spectra are only marked but not excluded Therefore it is recom mended to use only suitable spectra for external validation External Yalidation I Show Single Classifications i p Cancel Help Class 1 Browse Class 2 Browse Figure 9 43 External Validation dialog for a two class model default setting Show Single Classifications Check this option if a Classification report should be shown for each class in the model ClinProTools User Manual Version 2 2 9 55 Reference Part ClinProTools Menus Bruker Daltonik GmbH Class 1 Class 2 Class n Enter name and path of the validation spectra for class 1 class 2 class n Alterna tively you can select the respective spectra via a browser dialog For this click Browse of the respective class entry box navigate to the spectra and click OK OK Classifies the validation spectra and shows the results in the Validation report If defined additionally a Classification Validation report is shown for each class in the model separately 9 1 6 3 Save Classification Command The Save Classification command is used to save the classification result for the loaded spectra collection in an XML file with a s
273. s Each button is supplied with a tool tip showing a short description of its function Both toolbars are docked below the menu bar by default You can hide a toolbar and show it again using the General Toolbar resp View Tool bar command from the View menu A toolbar can be undocked by double clicking its slider and docked again by double clicking its title bar It can be moved by positioning the mouse cursor in the slider title bar and dragging the bar with the left mouse button held down to the desired position are eo Bl aa A ae Ul Ey HF CE ak Figure 5 9 General toolbar left and View toolbar right 5 1 6 Status Bar The status bar Figure 5 10 is docked at the bottom of the ClinProTools window You can hide the status bar and show it again using the Status Bar command from the View menu For Help press F1 CAP NUM X 6658 3 m z Y 14 71 arb u Loading Spectra Collection 50 0 Figure 5 10 Status bar In the left hand corner a short help text is displayed This corresponds to the cursor position and given actions The next boxes show if the caps lock CAP and the alphanumeric function NUM are activated In the right hand corner the progress of a running process is shown The boxes X and Y display the current x and y positions of the cursor if the cursor is in a data plotting view and Coordinates command for the respective window is active The data shown depends on the view the cursor is located in the process
274. s affected but the flatness of the baseline is also reduced This is especially of interest for mass ranges gt 20 kDa where the baseline correction might otherwise remove too much from broad overlapping peaks Baseline Flatness For Convex Hull Baseline selected enter the flatness of the baseline This parameter influences the number of parabola used to explain the baseline and the flatness of the resulting spectrum which is obtained by subtracting the baseline from the spectrum The larger the flatness value the finer the baseline will approach the spectrum 9 32 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Reference Part ClinProTools Menus In Mass Range define the parameters for the mass range filter Section 6 1 3 1 If you want to limit the mass range specify a minimal and a maximal mass Otherwise define a mass range that is larger than the experimental mass range which should be the case if you keep the default values Minimal Mass Enter the minimal mass of the mass range All masses below this value will be cut before loading the spectra Maximal Mass Enter the maximal mass of the mass range All masses above this value will be cut before loading the spectra Maximal Mass must be at least two times larger than Minimal Mass In Savitsky Golay Smoothing define the parameters for the smoothing filter Section 6 1 3 1 Enable Check this option if the Savitsky Golay smoothing filter should be enabled
275. s zoomed in You can also use the mouse wheel to zoom on the axes and in the Spectra View around the position of the mouse cursor to 65 or 150 respectively depending on the direction Expanding contracting and displacing axes The scaling of the x axis and y axis of a view can be changed using the mouse The scaling cursor is displayed when the mouse cursor is positioned on below the x axis or on left to the y axis it To expand or contract an axis drag the scaling cursor with the right mouse button held down right upwards to expand left downwards to contract To displace an axis drag the scaling cursor with the left mouse button held down in the desired direction Alternatively you can use the various Scaling com mands available in the view s context menu 5 12 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH ClinProTools User Interface You can also use the mouse wheel to displace an axis when the Shift key or the Ctrl key respectively is simultaneously held down Using the Shift key displaces the axis by 15 using the Ctrl key shifts the axis by 90 of window extent Undoing Redoing display range changes ClinProTools stacks the zooming operations for each view separately You can use the Undo Zoom and Redo Zoom commands from the View menu to undo redo the last done undone display range change in the focused view You can reset the Spectra Gel 2D Peak Distribution or Single Peak Variance View to f
276. s Spectra Preparation dialog as desired If spectra are loaded and you change parameters that affect spectra preparation during loading the views become cleared In this case close all spectra and open the classes again 2 From the Data Preparation menu select Recalibration 7 6 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Workflows in Detail 7 1 5 Setting up the Average Peak List The average peak list determines the peaks to be calculated in the individual spectra It collects all peaks that were picked on either the total average spectrum or the single spectra as well as manually edited peaks Each peak gets an index number and is defined by its m z value and integration region 7 1 5 1 Calculating the Average Peak List The average peak list can be calculated by automatically picking peaks on either the total average spectrum or the single spectra Automatic calculation is based on the peak picking parameters defined in the Settings Peak Calculation dialog Section 9 1 4 2 The picked peaks are indicated by gray integration regions in the Spectra View A later manual editing of the found peaks is possible Section 7 1 5 2 The average peak list calculation workflow can be started manually using the Average Peak List Calculation command from the Data Preparation menu The workflow will be run automatically if a workflow that requires an average peak list being calculated is launched without the average peak list c
277. s and coeffi cients Double clicking that file displays the respective data The file is overwritten on each PCA run 7 5 2 2 Influence Plot The Influence plot Figure 5 15 shows which influence a spectrum has on the current PCA model It is based on a certain number of most variant PCs you have to specify to set up the plot via the Influence command in the Plots menu Each data point in this plot represents a spectrum included in the PCA model The Influence plot is a diagnostic tool for the identification of outliers The vertical axis is a measure how far away a spectrum is from the model space distance to model The horizontal axis is a measure how far away a spectrum is from the model center after being projected into the model space and thus of the leverage of a spectrum to the model Strong outliers have high leverage on the model i e strong power to pull the PCA model toward themselves and may consume one PC just because of their existence The term leverage derives from the Archimedean principle that anything can be lifted out of balance as long as the lifter has a long enough lever 7 22 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Workflows in Detail The display of the Influence plot can be changed by zooming and panning operations data points can be marked with spectrum information and the content of the plot can be copied to the clipboard like described for the Scores and Loadings plots Section 7
278. s instead of blue ones Figure 4 2 Models are not saved automatically If you want to save a model select it from the list and save it using the Save Model As command from the Model List View Save context menu or ClinProTools User Manual Version 2 2 4 5 Getting Started with ClinProTools Bruker Daltonik GmbH ked M ClinProTools Class 1 Normal Class 2 Spil F ta odi Ea foc 144 JE E He al New Calc Cance Show Classify Load Clear All Model List Save Validate Model Name Algorithm Cross Validation Recognition Capability State Modell GA 100 00 100 00 Calculated Model2 SVM 100 00 100 00 Calculated Model3 SNN 100 00 100 00 Calculated Model4 ac 100 00 100 00 Calculated ba fae Ga lS a Se UE oo F ClinProt Model DER Datei Bearbeiten Ansicht Favoriten Extras ay 2 ClinProt Model List ay I gt x El EA pe Search She Favortes O A 7 A j x Search Favorites d B a r K s C BDAL ClinProTools_2_2 Files ClinProtMadelo002 xml s E C BDAL ClinProTools_2_2 Files ClinProtModelList0001 xml e ClinProt Model ClinProt Model List ClinProTools Version 2 2 build 38 BRUKER Name Modell Date Time 2007 05 16T08 56 33 826 02 00 Validation GUID 4dabaea2 1047 41c4 ad21 1b818aeb582c Name Algo E ES ClinProTools 5 o build 38 Rec Max Max Auto Version
279. s of cross valida ClinProTools User Manual Version 2 2 7 13 Workflows in Detail Bruker Daltonik GmbH tion and recognition capability if calculated To show a single model 1 In the Model List View right click the model you want to show and select Show Model or click __ Show 7 2 1 6 Showing All Models in the Model List You can show all models currently in the model list in the Model List report Section 8 1 1 5 This includes all models of any state with corresponding parameters The data shown depends on the models current state To show all models in the model list 1 From the Reports menu select Model List or click _Model List 7 2 1 7 Saving a Model If you want to keep a calculated model you can save it in an XML file with a specified name This allows loading the model again e g for external validation or classification of test spectra To save a model 1 In the Model List View right click the model you want to save and select Save Model As or click 22E 2 In the Save Model dialog specify the file name and target folder The ClinProt Models folder is the default storage location 3 Click Save 4 If you have selected an existing name answer the confirmation request to over write the file 7 2 1 8 Removing a Single or All Models from the Model List You can remove a model from the model list if you do not want to have it there any longer or clear the complete model list at once This unloa
280. s the clusters calculated from the spectra classes and the distances among the single clusters Depending on the clustering parameter settings used the created dendrogram shows either the full tree of spectra with without spectra paths or is limited to a specified number of clusters the spectra were assigned to The corresponding data is stored in the follwing files in the ClinProTools folder e ClinProtClustering xml contains a list of classes with all the spectra paths of the spectra belonging to the classes only if number of classes is limited This file will display automatically after the clustering is completed if the corresponding option is set in the Unsupervised Clustering dialog and the Create Full Tree option is unchecked ClinProtClustering tree xml exports all the nodes with distances of the hierarchical clustering in the form of a linkage list ClinProtClustering tree2 xml exports the XML tree each node is represented by a node element which contains two sub node elements 7 24 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Reporting Data 8 REPORTING DATA ClinProTools offers various types of reports to report specific spectra peak statistic model validation or classification data as well as error information Open reports can be saved and printed Graphics from views plots can be printed or copied to the clipboard Peak list data can be exported to XML or CART format 8 1 Creating ClinProtTools
281. sed for the peaks of the respective class in the 1D peak distribution Section 9 1 3 6 8 arb u Spiked DatalSpiked 0_N10_1SLinVfic arb u Spiked DatalSpiked0_N10_1SLin fid 2636 mz 2636 miz Figure 9 14 Standard box plot left and modified box plot right indicating the outliers three by red crosses and three green circles not belonging to the 95 of values inside the whiskers ClinProTools User Manual Version 2 2 9 19 Reference Part ClinProTools Menus Bruker Daltonik GmbH 9 1 3 6 11 Spectra View gt Peak Markers Command The Peak Markers command shows hides the peak markers in the Spectra View Figure 9 15 A black arrow marker v at the top of the view indicates the peak s for which corresponding data is shown in the 2D Peak Distribution ROC Curve or Single Peak Variance View The peak markers are hidden by default arb u D Data Files ClinProTools ClinProTools Test Data EDTA Run0hiSample0_G11_1SLinitid 1400 1500 1600 1700 1800 1900 miz Figure 9 15 Peak markers indicating in the Spectra View the two peaks selected in the 2D Peak Distribution View 9 1 3 7 Gel Stack View Popup Command Pointing to Gel Stack View offers the following commands Figure 9 16 Gel Stack View A v Class Names v Current Spectrum Marker v Colored Spectrum State Excluded Spectra v Group Separators Follow Spectra View Mass Range Figure 9 16 Gel Stack View submenu Command Used to Class Names Sh
282. splayed in a separate window via the corresponding command of the Plots menu To view the data in a Scores plot or Loadings plot in detail you can change the dis play of the plot by zooming panning and rotating operations via the Zoom Pan and Rotate commands of the View menu If you want to know which spectrum corresponds to a data point in a Scores plot or which m z value corresponds to a data point in a Loadings plot just click the desired point with the left mouse button This marks the selected data point with file a awl resp m z 8513 description Clicking a marked data point again removes the information from that point To un mark data points the Mark Data Points command of the View menu must be active default setting e The scores and loadings of PC1 PC2 and PC3 are plotted against each other by default If you want to view the data of another PC set you can change the PC selection via the PCs command of the PC menu Each PC change updates all plots in the PCA main window accordingly but previously set up single plot windows remain unchanged A graphic of the content of the PCA main window or a single Scores plot or Loadings plot window can be copied to the clipboard via the Copy command of the Edit menu This allows pasting the graphic into an appropriate application e All PCA data generated during the current PCA are stored as ClinProtPCA xmiI file in the ClinProTools folder This includes the calculated variances score
283. starts After clustering is completed the Dendrogram window opens displaying the clustering result Section 7 6 2 Shortcut Button 9 1 8 Reports Menu The Reports menu offers the following commands Figure 9 47 Spectra List Peak Statistic Correlation Matrix Settings Statistic Figure 9 47 Reports menu Command Used to Spectra List Show the Spectra List report Peak Statistic Show the Peak Statistic report Correlation Matrix Define correlation parameters and show the Correlation Matrix report Model List Show the Model List report Settings Statistic Define settings for calculating peak statistic and showing certain statistical data in the Spectra View 9 60 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Reference Part ClinProTools Menus 9 1 8 1 Spectra List Command The Spectra List command creates and shows the Spectra List report Section 8 1 1 1 and stores the data as ClinProtSpectraList number xml file The Spectra List report lists all loaded spectra with corresponding data 9 1 8 2 Peak Statistic Command The Peak Statistic command creates and shows the Peak Statistic report Section 8 1 1 2 and stores the data as ClinProtStatistic fnumber xml file The calculation is based on the current statistic settings Section 9 1 8 5 The Peak Statistic report lists all picked peaks with corresponding data The 2D Peak Distribution View displays the first two peaks of the selected
284. statistical sort order by default if currently active If the spectra recalibration average peak list calculation and or peak calculation workflow s have not been performed yet the respective workflow s will be automatically run before calculating the peak statistic Note By default the peak statistic is calculated using the same settings as the peak selection However you can use differing settings if desired Shortcut Button as 9 1 8 3 Correlation Matrix Command The Correlation Matrix command is used to calculate a correlation analysis Section 6 4 2 1 which compares each peak in the peak list to each other peak and to create and show the corresponding Correlation Matrix report Section 8 1 1 3 The correlation analysis can be calculated over either all classes or only a specified one The result is stored as ClinProtCorrelationMatrix number xml file The settings for correlation matrix setup are stored in the SettingsGeneral xml file which is updated each time you change the correlation settings or any other settings saved to this file Note Resetting the general settings also resets the current correlation settings The command opens the Correlation Matrix dialog Figure 9 48 to specify how to set up the correlation matrix and to start correlation analysis If the spectra recalibration average peak list calculation and or peak calculation workflows have not been per formed yet the respective workflow s will be automaticall
285. sted person is truly diseased sensitivity true positives true positives false negatives sensitivity TP TP FN In our example validation sensitivity 24 24 3 sensitivity 88 9 Specificity The specificity of such a test is the probability that the test has a negative outcome when the tested person is truly not diseased specificity true negatives true negatives false positives specificity TN TN FP In our example validation specificity 19 19 7 specificity 73 1 Sensitivity alone does not tell us all about the test because a 100 sensitivity can trivially be achieved by labeling all test cases positive and a 100 specificity can trivi ally be achieved by labeling all test cases negative However in the first case the ClinProTools User Manual Version 2 2 6 43 Basics Bruker Daltonik GmbH specificity would be zero and in the second case the sensitivity would be zero respectively A test with a high sensitivity has fewer Type II errors a test with a high specificity has fewer Type errors For explanation of Type and Type Il errors please refer to the Glossary Appendix A 2 Note e Sensitivity true positives true positives false negatives e Specificity true negatives true negatives false positives e Positive prediction true positives true positives false positives e Negative prediction true negatives true negatives false negativ
286. submenu Command Used to Single Spectra Show Hide the single spectra with one single spectrum displayed at a time All Single Spectra Show Hide the simultaneous overlaid display of all single spectra Total Average Show Hide the total average spectrum Spectrum Average Spectra Show Hide the average spectra for a specified class es Noise Spectrum Show Hide the noise spectrum Integration Regions Show Hide the integration regions of picked peaks Average amp StdDev Show Hide the class averages of peak area intensity with standard deviation Peak Distribution Show Hide the peak distribution plot with respect to peak area intensity Box amp Whiskers Show Hide the box amp whiskers plots for the peak area intensity per class ClinProTools User Manual Version 2 2 9 13 Reference Part ClinProTools Menus Bruker Daltonik GmbH Command Used to Outliers for Box amp Show Hide the outliers for box amp whiskers plots for the peak Whiskers area intensity per class Peak Markers Show Hide the markers for the peak s selected in the 2D Peak Distribution ROC Curve or Single Peak Variance View 9 1 3 6 1 Spectra View gt Single Spectra Command The Single Spectra command shows hides single spectra of the loaded classes in the Spectra View with displaying only one single spectrum at a time All single spectra within a class are displayed in the same predefined class color excluded spectra are shown with a darker colo
287. t 9 62 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Reference Part ClinProTools Menus 9 1 8 4 Model List Command The Model List command creates and shows the Model List report Section 8 1 1 6 and stores the data as ClinProtModelList number xml file The Model List report lists all models currently contained in the Model List View Shortcut Button Model List 9 1 8 5 Settings Statistic Command The Settings Statistic command is used to define the settings to be used to calculate peak statistic By default the same settings are used as are defined for the peak selec tion Section 9 1 5 1 but you can define differing settings e g to set up a Peak Statistic report sorted by mass value Furthermore you can limit the number of peaks for which statistical data average with standard deviation peak distribution and box and whiskers is displayed in the Spectra and Single Peak Variance views when corresponding View menu commands are active The statistic settings are stored in the SettingsGeneral xml file which is updated each time you change the statistic or any other settings saved to this file The command opens the Settings Statistic dialog Figure 9 49 Note Resetting the general settings also resets the current statistic settings Settings Statistic Use Selection Sort Mode V From Settings Peak i Cancel Peaks to Show in Views Help r Peaks to Show Peak Statistic Sort Mod
288. t E Edit menu Edit menu MATLAB Edit Model Name command Edit Peak command Error report Error treatment Exclude Peak command Exclude Spectrum command Excluded Spectra command Excluding Peak Spectrum Exit command Explained variance PCA 9 73 9 73 9 73 9 73 l 2 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Index Exporting peak list CART format 8 14 XML format 8 14 External shutdown of ClinProTools 10 1 External validation 6 24 7 15 External Validation command 9 55 F File location in ClinProTools 4 2 File menu 9 1 Filters modifying spectra 6 8 Filters selecting spectra 6 9 Follow Spectra View Mass Range command 9 23 Force Peak into Model command 9 77 Forcing peak into model 7 13 G Gel View 5 4 Gel View context menu 9 68 Gel Stack View 5 3 Gel Stack View popup command 9 20 General settings Defining 4 2 Resetting 4 2 General Settings command 9 6 General settings parameters 9 6 General toolbar 5 10 General Toolbar command 9 12 Genetic Algorithm 6 12 6 13 Genetic Algorithm parameters 9 45 Glossary A 6 Grid command 9 77 Group separators 9 23 Group Separators command 9 23 H Help menu 9 65 Help Topics command 9 66 l Include Peak command 9 76 Include Spectrum command 9 10 Including Peak 7 9 Spectrum 7 5 Influence command MATLAB 9 84 Influence plot PCA 7 22 Influence window PCA 5 16 Info Loaded Classes command 9 4 Installation notes 2 2 Installing ClinProTo
289. t for example be necessary in the case ClinProTools is used for classi fication by the Bruker flexlmaging software In this mode the spectra to be classified are neither displayed in ClinProTools nor kept in the memory any number of spectra can be classified at a time After the classification a saving dialog pops up automa tically to store the classification result in an XML file The Classification report can be created on demand however it is not recommended to display big classifications because the browser used for display might take a long time to process the XML file with style sheet Large XML files with style sheet should better be opened in Excel In both modes the software holds the classification result as long as the classification is not closed ClinProTools User Manual Version 2 2 6 25 Basics Bruker Daltonik GmbH 6 4 Statistics in ClinProTools ClinProTools supports various statistical tests and methods which can be applied to the prepared spectra data A short introduction to each test method is given in this sec tion In addition some remarks to statistical problems with MS data have to be made 6 4 1 Statistical Tests ClinProTools offers various statistical tests With respect to the constraints for the individual tests we can differ between tests expecting normal distribution and distribu tion free tests The t test and ANOVA test are tests expecting normal distribution of the underlying data and are for
290. t in the Loadings A window Loadings B Display the middle 2D Loadings plot in the Loadings B window Loadings C Display the right 2D Loadings plot in the Loadings C window 9 3 3 1 Variance Command The Variance command displays the Variance plot of the PCA in the Variance window 9 3 3 2 Influence Command The Influence command is used to display the Influence plot for a specified number of PCs The command opens the Influence dialog Figure 9 66 to set the number of PCs 9 84 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Reference Part MATLAB Based Menus to be concerned The number of PCs needed to explain 95 of the variance in the spectra set is suggested by default Clicking OK creates the corresponding Influence plot and shows it in the Influence window mm Influence Number of PCs Figure 9 66 Influence dialog 9 3 4 PC Menu The PC menu of the PCA main window offers the following command Command Usedto PCs Select the PCs for which corresponding data should be displayed in Scores plots and Loadings plots 9 3 4 1 PCs Command The PCs command is used to select the PCs for which you want to view data in the Scores plots and Loadings plots It opens the PCs dialog Figure 9 67 to enter the desired PC numbers Clicking OK updates the data in the plots of the PCA main window accordingly plots previously set up in a separate window remain unchanged Figure 9 67 PCs
291. t the smallest deviation has a rank of 1 Tied scores are assigned for a mean rank The sums for the ranks of scores with positive and negative deviations from the central point are then calculated separately A value S is defined as the smaller of these two rank sums S is then compared to a table of all possible distributions of ranks to calculate p the statistical probability of attaining S from a population of scores that is symmetrically distributed around the central point As the number of used scores n increases the distribution of all possible ranks S tends towards the z distribution so for an n of greater than 10 this distribution is used to calculate p This test assumes that the compared sample sets originate at least from a common distribution For details please refer to F Wilcoxon Individual Comparisons by Ranking Methods Biometrics 1 pp 80 83 1945 6 4 1 4 Kruskal Wallis Test In statistics the Kruskal Wallis one way analysis of variance by ranks is a non parame tric method Unlike the analogous one way analysis of variance the Kruskal Wallis test does not assume a normal population This like many non parametric tests uses the ranks of the data rather than their raw values to calculate the statistic Since this test does not make a distributional assumption it is not as powerful as the ANOVA test The hypotheses for the comparison of two independent groups are e HO null hypothesis The samples come from identica
292. ta follow a specific distribution we already know a lot about the data and their behavior Therefore tests have been developed which give very nice powerful test procedures as long as the data are nearly e g normally distributed The t test and the ANOVA test are tests of this kind and their results are only valid from a formal point of view if the underlying data fits the normal distribution constraint Applying these tests on non normally distributed data will distort the test results To constraint C3 This constraint is very important and common for all tests In general a large number of disjunctive no multiple measurements or duplicates samples improves the power of the test and make thus the results more reliable A small sample size on the other hand increases the probability for a wrong test decision To constraint C4 In clinical proteomics we search for potentially interesting peaks or masses which are capable to separate different classes e g cancer control In contrast to the common assumption the more the better a large number of features does not necessarily improve the performance of the search for such candidates This is mainly because many of the features are not important for the underlying classification question and hence are in some sense noise To overcome this it is important to have a good parameterization for the peak detection e g S N threshold to pick only peaks that have a good S N ratio and are likel
293. ted peak from the peak list Change the integration region of the selected peak Display in the ROC Curve view the ROC curve for the selected peak resp in the Single Peak Variance View the variance for this peak Calculate per peak correlation analysis for selected peak ClinProTools User Manual Version 2 2 9 67 Reference Part ClinProTools Context Menus Bruker Daltonik GmbH 9 2 2 Gel View Context Menu The Gel View context menu offers the following commands Command Used to Coordinates Show Hide the display of cursor coordinates in the status bar Grid Show Hide the grid in the view Scaling Pop up scaling commands for the view Zooming Activate Deactivate the zoom in mode in the view Undo Zoom Same as Undo Zoom command from View menu Redo Zoom Same as Redo Zoom command from View menu Distance Switch the view to distance measurement mode Display Type Pop up commands for toggling between Gel and Stack views Display Mode Pop up display modes for the view Exclude Include Same as Exclude Include Spectrum command from Edit menu Spectrum Right clicking the Gel View s color bar opens a context menu containing the same com mands as offered by the Display Mode command 9 2 3 Stack View Context Menu The Stack View context menu offers the following commands Command Used to Scaling Pop up scaling commands for the view Display Type Pop up commands for toggling between Gel and Stack views Whitewash Switc
294. tep protocols for bead combination 3 2 Sample Preparation For MALDI TOF MS analysis it is recommended to prepare the samples on Bruker 384 MTP AnchorChip targets with an optimal anchor diameter of 600 um 209513 In general target preparation can be performed with a number of different matrices e g according to the mass range of interest and based on different protocols For details about MALDI TOF MS target preparation please refer to the AnchorChip manual ver sion 2 2 In the following a protocol is described which has been optimized for special clinical proteomic approaches to gain profile spectra in the mass range from approx 1000 to 10000 Da Please keep in mind that high reproducibility of results is significantly depending on reproducibility of sample preparation in all different steps starting from collecting and storing the samples and ending with target preparation and MALDI TOF analysis ClinProTools User Manual Version 2 2 3 1 Data Acquisition for Clinical Proteomics Bruker Daltonik GmbH Target preparation protocol for profiling samples Matrix for the mass range 1 20 kDa Matrix solution a cyano 4 hydroxycinnamic acid HCCA 201344 201072 0 3 g l in ethanol acetone 2 1 daily prepared Take 1 ul of the sample purified and directly eluted from the magnetic beads according to the protocol and mix it thoroughly with 10 ul of the matrix solution Subsequently 0 5 to 1 ul of the mixture should
295. the mouse cursor into the desired plot Position the zoom in cursor S at the desired start point and draw it holding the left mouse button pressed to the desired end point On releasing the mouse button the enclosed area is zoomed in The zooming in steps you perform in a plot are stacked for each plot separately which allows stepwise zooming out of the plots Double clicking with the left mouse button restores the plot s original view When the command is active right clicking a 2D plot pops up a context menu offering the following commands Command Used to Zoom Out Stepwise zoom out the selected plot if it was zoomed in before Reset to Original View Restore the plot s original view Zoom Options Unconstrained Zoom Allow unconstrained zooming Horizontal Zoom Allow zooming in horizontal direction only Vertical Zoom Allow zooming in vertical direction only 9 3 2 3 Pan Command The Pan command switches to the pan mode This mode allows moving all data points within a 2D plot as an entity in any direction This e g enables longer descriptions attached to data points to be read To pan a plot move the mouse cursor into the desired plot so that it changes into the pan cursor Click the plot with the left mouse button and while holding the mouse button pressed shift the data points in the desired direction Double clicking with the left mouse button restores the plot s original view When the command is active right clicking a 2D p
296. the relation between the peaks Do peak areas vary independent of each other or is there a correlation between the variation of the peak areas The correlation coefficient cc can be between 1 and 1 A cc of 1 means that the areas of the two peaks have a perfect positive correlation and go up and down in the same way If the cc is 1 the two peaks are perfectly negative correlated If the intensity of the first peak is above the mean level in one spectrum the other peak will be below the mean level and vice versa Smaller absolute values of the cc indicate that the peak areas are not correlated and vary in an independent way ClinProTools uses a color code ranging from red cc 1 to blue cc 1 to highlight different cc ranges in the matrix and thus allows quickly detecting highly correlated peak pairs The correlation matrix list can be used to identify peaks that show concordances or discordances If two peaks behave similar i e their cc is close to 1 and if they are relevant for the classification task it is quite common that only one of these peaks is incorporated in the model In this case alternative peaks which behave similar to the one available in the model can be identified using the correlation matrix list In a correlation matrix it is also possible to identify groups of peaks which are highly correlated To group the peaks of the correlation matrix we start with one peak and add all peaks to the first group which hav
297. the spectra of the third loaded class class 3 above the class 2 spectra etc Each class is separated from the next loaded one by a horizontal line Classes used in model generation are separated by thin lines a class used in classification is separated from the last loaded model generation class by a thicker line The class names Section 9 1 3 7 1 consisting of path and folder name of the respective class are shown by default The color box in front of the class name indicates the color the single spectra contributing this class are displayed with in the Spectra View 12 class colors are predefined in the system after 12 classes repetition of colors will occur 5 4 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH ClinProTools User Interface The current spectrum marker Section 9 1 3 7 2 marks the spectrum currently shown in the Spectra View When spectra from multiple measurements are loaded and spec tra grouping is enabled dashed group separators Section 9 1 3 7 5 are shown by default which separate spectra originating from the same spot Like the Spectra View the Gel View allows manual exclusion inclusion of unprocessed spectra Manually as well as automatically excluded spectra are highlighted by default using colored spectrum states concerning the reason of exclusion Section 9 1 3 7 3 Excluded spectra can be hidden from the Gel View Section 9 1 3 7 4 5 1 2 2 Stack View The Stack View Figure 5 4 disp
298. timal separating hyperplane the solid line repre sents the obtained optimal line or with more than two peaks hyperplane The advantage of the SVM is that it is quite fast in the determination of the peak ranking and a well formal established pattern recognition tool yielding good results The drawback is that it calculates its solution including all peaks as a whole and is by principle a two class approach the multi class solution is obtained by a wrapper method which may lead only a near optimal solution since you cannot guarantee to find the best combination if you do not test all of them If we have classification problems with a large number of peaks we want to know which peaks separate best Therefore the solution obtained from the SVM can be ana ClinProTools User Manual Version 2 2 6 15 Basics Bruker Daltonik GmbH lyzed in more detail to get a ranking of the contributing peaks Peaks which have good separation properties are more important for the SVM solution than peaks which do not separate well How the SVM works If we have more than two classes we split the data into a class containing all data points from the current considered class and a rest class which contains the remaining data points A penalty term C is determined automatically from the data to limit the structural risk of misclassifications and the formal optimization problem is defined The optimization problem is solved using a quadratic problem sol
299. tion 0_A14_15Lin D Data Files ClinProTools ClinProTools Test DatajEDTA Run 06h 5 UKE E06 1 EDTA_6h Plasma AF ProcessStep Sample PreparationMe HIC C8 lt Figure 9 64 Spectrum Information dialog 9 2 9 26 Whitewash Command The Whitewash command switches the Stack View to whitewash mode In this mode the plot is structured finer due to resolving overlying structures All spectra are drawn in black color thus their class membership is not shown Figure 9 65 9 80 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Reference Part ClinProTools Context Menus arb u 150 100 50 2000 6000 8000 miz Figure 9 65 Stack View in whitewash mode 9 2 9 27 Zooming Command The Zooming command activates deactivates the Zoom for the Spectra Gel 2D Peak Distribution or Single Peak Variance View to zoom in the selected range Section 5 1 7 2 When this command is active for a view the Zoom in cursor A is displayed when you move the mouse in the respective view Otherwise the mouse cursor is displayed ClinProTools User Manual Version 2 2 9 81 Reference Part MATLAB Based Menus Bruker Daltonik GmbH 9 3 MATLAB Based Menus The PCA windows and the Dendrogram window provide own menus originating from the external MATLAB tool integrated in ClinProTools The menus and commands available depend on the particular window 9 3 1 Edit Menu The Edit menu of a PCA window or the Dendrogram windo
300. tion about the variability within model generation classes and thus the homogeneity heterogeneity of a spectra set a PCA can be carried out within Clin ProTools In the context of PCA the separation into classes is ignored i e all data is treated as one group It is also possible to apply PCA to a single loaded class only e g to detect subgroups or outliers within the model generation class PCA can be performed on grouped spectra too The PCA is carried out by an external MATLAB software tool which is started automatically within ClinProTools 7 5 1 Calculating a PCA A PCA is calculated on all non excluded spectra in the loaded spectra set s and requires two valid spectra with three peaks being available at least The PCA workflow automatically runs the spectra recalibration average peak list calculation and or peak calculation workflows if these have still not been performed when launching PCA calculation After the PCA is completed the PCA main window opens displaying the results To calculate a PCA 1 Open the spectra set s on which you want to calculate a PCA If certain spectra should not be included in PCA exclude them 2 3 From the Statistical Analysis menu select PCA or click i 4 In the PCA dialog check whether normalized data should be used Click OK to start PCA If required the spectra recalibration average peak list calculation and or peak calculation workflows will be run prior to starting PCA 5 View the r
301. tly loaded follow the instructions in the appearing message on how to proceed ClinProTools User Manual Version 2 2 7 3 Workflows in Detail Bruker Daltonik GmbH To reset the current data preparation settings to the defaults 1 From the Data Preparation menu select Reset Settings Data Preparation 2 Confirm the appearing request to reset the current settings to the defaults 3 If spectra are currently loaded follow the instructions in the appearing message on how to proceed 7 1 2 Loading Spectra in ClinProTools To generate a model or calculate statistics one or several classes have to be loaded ClinProTools loads all spectra in a folder and its subfolders recursively as one class ClinProTools supports loading spectra of the X Mass BAF und ASCII file formats For loading ASCII files Appendix A 4 the Null Spectra Exclusion filter Section 6 1 3 2 in the Settings Spectra Preparation dialog has to be disabled For model generation two classes must be loaded at least Single classes can be loaded for peak statistic operations performing PCA or unsupervised clustering You have to load all classes that should be included in model generation or statistic opera tions before you start any data processing e g recalibration peak calculation A later loading of additional classes is not possible without starting complete data preparation and thus spectra loading again There are two ways to load spectra in ClinProTools
302. to Noise Threshold or by giving the Maximal Peak Number found in the Settings Peak Calculation dialog Furthermore peaks which do not exceed a certain percentage of the largest peak Relative Threshold Base Peak can be excluded The integration regions of the picked peaks can be displayed in the Spectra View Section 9 1 3 6 6 6 1 1 5 2 Peak Picking on the Single Spectra In the single spectra peak picking approach an overall average peak list is calculated over all spectra of all classes by an automatic combination of multiple peak lists Thereby the peak lists are determined by application of a peak picking procedure for each sample All single peak lists are merged together such that a large list of peaks is obtained which in principle includes duplicates that are only distinct by a small error or mass shift The procedure starts with the average peak list obtained in the standard approach This average peak list is used to screen out peaks which are already very common These peaks immediately become part of the final overall average peak list On the remaining set of peaks a clustering is applied Martinez et al 1993 which is combined with the approach published in DeSieno 1988 to obtain a clustering of the peaks Thereby the number of clusters is determined in accordance to an overestimate of the number of peaks in the given list of peaks The clustering forces peaks to be in a cluster which are very close to each other by means of
303. tomize settings for that filter The dialog can only be edited if Enable is checked 9 34 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Reference Part ClinProTools Menus Settings Adduct Polymer Spectra Exclusion IS Atomic Weight Da Mass Tolerance 21 98 1 Cancel 23 3 Help 38 09 44 58 1 1 1 1 Figure 9 30 Settings Adduct Polymer Spectra Exclusion dialog default setting Name lists the adducts polymers the filter should search the spectra for Sodium Na magnesium Mg potassium K polyethylenglycol PEG and polypropylene glycol PPG are contained by default You can add new adducts polymers to this list and change or remove existing adducts polymers Atomic Weight Da lists the corresponding atomic weights of the adducts poly mers Mass Tolerance Da lists the corresponding mass tolerance allowed for detecting the adduct polymer peak Add Edit allows adding a new resp editing the selected adduct polymer The button opens the Adduct Polymer Property dialog Figure 9 31 to add a new adduct polymer to be searched for by the filter or to edit the selected one with respect to atomic weight and or mass tolerance Do the desired entries changes and click OK to add the new adduct polymer to the Settings Adduct Polymer Spectra Exclusion dialog resp change the selected one Adduct Polymer Property Adduct Polymer Property Atomic Weight Da Mass Tolerance Da Atomic Weight Da
304. tra of the loaded model generation classes the calculated average and noise spectra and specific peak statis tics The x axis records the m z value the y axis the peak intensity in arbitrary units The statistical plots are drawn on a unique scale independent of the peak intensity scale The kind of spectra and peak statistics displayed depends on the current pro cessing state and the corresponding View menu settings Section 9 1 3 6 Single spectra are displayed by default with one single spectrum shown at a time Section 9 1 3 6 1 Alternatively you can show all single spectra in an overlay spectra plot Section 9 1 3 6 2 The path and name of the current spectrum are indicated in the top right corner of the view All single spectra of one class display in the same color that is indicated in the Gel View To show another single spectrum of the same or another class use the scroll bar or click the spectrum in the Gel View After spectra recalibration the total average spectrum Section 9 1 3 6 3 is shown by default in gray color In addition you can show class average spectra Section 9 1 3 6 4 displayed with a darker color than the corresponding single spectra e g red gt dark red and or the noise spectrum in orange Section 9 1 3 6 5 After peak calculation and peak selection all picked peaks are marked by colored integration regions Section 9 1 3 6 6 by default included peaks with blue excluded peaks with gray bars After mod
305. tries to identify some characteristic spectra for each class which are named prototypes and which could be somehow considered as prototypical samples of that class Support Vector Machine The Support Vector Machine SVM is an algorithm for the determination of optimal separating planes between different data classes It uses formal approaches from optimization theory to separate the given data sets Upon the obtained planes a peak ranking can be calculated in a second step Test data Test data are spectra to be classified by the software using a model containing peak patterns generated by the model generation process Type I and Type Il errors Type error and Type II error are common measurements in statistical testing and described in the following table Expected decision Reality Positive Glass Machine Positive Class True positive correct False negative type error Test decision false decision type 1 decision Negative Class Negative Class False positive type II True negative correct error false decision decision type II Unsupervised clustering The hierarchical clustering approach is used as an unsupervised clustering method in ClinProTools A hierarchy of clusters is generated which is represented in the form of a dendrogram a tree Validation After generation of a model this needs to be validated Validation in case of a two class scenario yields estimates of
306. trum Mark Do not mark in the 2D Peak Distribution View the data Marker point corresponding to the current spectrum 9 26 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Reference Part ClinProTools Menus 9 1 3 8 5 1 Peak Statistics View gt 2D Options gt Select Peaks Command The Select Peaks command is used to change the current peak selection in the 2D Peak Distribution View By default the first two peaks of the current statistic sort order set via the Settings Statistic command Section 9 1 8 5 are displayed The horizontal axis plots the first the vertical axis the second peak The command opens the Peak Distribution dialog Figure 9 24 to select two peak indices for displaying the corresponding peak distribution Peak Distribution Select Peak Indices for Peak Distribution Display Cancel V Use First Two Peaks Vertical Help Hao TE Horizontal Figure 9 24 Peak Distribution dialog Use First Two Peaks Check this option if the first to peaks of the current statistic sort order should be dis played Uncheck this option if you want to display the peak distribution for another pair of peaks which you have to specify in Vertical and Horizontal Vertical Horizontal Enter the index of the peak to be displayed on the vertical axis Horizontal Enter the index of the peak to be displayed on the horizontal axis OK Updates the 2D Peak Distribution View for the now selected peaks 9 1 3 8 5
307. tter cases however the current peak calculation will be reset which is indicated in that the integration regions of all peaks change to gray color This requires running the peak calculation resp model generation workflow again The command displays the distance cursor Figure 9 60 Its two vertical lines mark the current limits of the integration region of the selected peak Move the cursor lines as described with the Distance command Section 9 2 9 10 to the new positions where the peak should start and end and click the right mouse button Confirm the appearing request to change the integration region of the selected peaks as stated Figure 9 61 This changes the integration region of the peak accordingly ClinProTools User Manual Version 2 2 9 75 Reference Part ClinProTools Context Menus Bruker Daltonik GmbH 2780 miz 2780 miz 2780 miz 2780 miz Figure 9 60 Changing the integration region of a peak using the Distance cursor Mass Range Change Integration Limits of Peak 10 to 2771 86 and 2787 97 Da Cancel Figure 9 61 Mass Range dialog to confirm changing of integration limits 9 2 9 13 Exclude Include Peak N Command The Exclude Include Peak n command excludes or includes respectively the selected peak in model generation Peaks can be excluded included after running the peak calculation workflow All included peaks have a blue integration region in the Spectra View all excluded peaks a gray one
308. tting a running number e g ClinProtSpectra0001 xml ClinProtSpectra 0002 xml ClinProtStatisticO001 xml ClinProtStatisticO002 xmi etc ClinProTools allows clearing all temporary files in the ClinProTools folder at once using the General Settings command from the File menu and clicking Clear Temporary XML Files If you do not want to delete all these files you can select and delete only the desired ones using e g the Microsoft Windows Explorer 4 3 ClinProTools General Settings ClinProTools allows defining certain general non algorithm settings for like file paths display of ClinProt xml files etc If you do not want to work with the defaults you can define own settings Changed settings can be reset to the defaults The ClinProTools general settings are saved in the SettingsGeneral xmI file This file is generated on ClinProTools start up and updated on each settings change If this file is not present when ClinProTools is started a new one with default values will be generated The SettingsGeneral xml file also collects the file open paths correlation and statistic settings Note Resetting the general settings is only possible when no spectra are loaded To view and change general settings for ClinProTools 1 From the File menu select General Settings 4 2 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Getting Started with ClinProTools 2 Inthe General Settings dialog change the default settings if des
309. two t test or k classes with k gt 2 ANOVA test The supported Wilcoxon test k 2 or Kruskal Wallis test k gt 2 do not depend on the normal distribution assumption and should be used for a more generic analysis The Anderson Darling test in the case of ClinProTools has been adapted to test for normal distributions Each of these tests calculates the so called p value For a detailed introduction in statistical tests including mathematical theory please refer to J M Chambers and T J Hastie Statistical Models in S Wadsworth amp Brooks Cole 1992 R E Walpole and R H Myers Probability and Statistics for Engineers and Scientists 5th ed Macmillan 1993 6 4 1 1 T Test A t test is a statistical hypothesis test in which the test statistic has a Student s t distribution if the null hypothesis is true In ClinProTools we consider the t test as a statistical test of the null hypothesis that the means of two normally distributed popula tions are equal All such tests are usually referred to as Student s t tests though strictly speaking that name should only be used if the variances of the two populations are also assumed to be equal the form of the test used when this assumption is dropped is sometimes called Welch s t test There are different versions of the t test depending on whether the two samples are e independent of each other e g individuals randomly assigned into two groups or e paired so that each membe
310. uired workflows recalibration average peak list calculation and or peak calculation are run according to the current settings In model generation all or only the selected peaks of the prepared non excluded spec tra of the loaded model generation classes are used The peaks that separate best between the loaded classes are searched for using the chosen classification algorithm and the algorithm related model parameters Cross validation is performed on the model and the recognition capability is calculated if both options are not deactivated The progress of model generation is shown in the Model List View s State column The results of cross validation and recognition capability calculation are entered in the model list after model generation has finished The model s state is changed to Calcu lated You can show the XML file of the calculated model Section 7 2 1 5 The calculation of new models can be canceled by clicking or Cancel Th model s state is set back to Added e To calculate a model 1 From the Model Generation menu select Calculate or click _Calculate This successively calculates all models of the state Added contained in the model list 7 2 1 5 Showing a Single Model You can show a calculated model in the Model report Section 8 1 1 6 This contains the model generation classes and all data preparation and model generation para meters that were used for setting up that model as well as the result
311. ull display of data by double clicking in the view If you want to reset only one axis you can double click it 5 1 7 3 Changing the Stack View s Orientation The orientation of the 3D Stack View is 30 with the base yielding approx one third of the window height by default You can quickly change the plot by combined dragging of axes with the mouse For this position the mouse in the Stack View to display the 3D cursor 2s and press the left mouse button This results in the current plot axes being displayed with bold black lines Figure 5 11 When dragging the mouse the orientation of the bold lines changes accordingly indicating the current positioning of three axes On releasing the mouse button the Stack View is updated immediately with redrawing all spectra Moving axes also allows changing the 3D into a 2D plot with displaying all spectra in list view Figure 5 12 LEELLE EH FFFFFFI EEEEE EL 4000 8000 mz 2000 4000 6000 8000 miz Figure 5 11 Stack View during dragging Figure 5 12 List view of spectra after axes bold black lines show the changing the 3D into a 2D current positions of axes plot ClinProTools User Manual Version 2 2 5 13 ClinProTools User Interface Bruker Daltonik GmbH 5 1 7 4 Resetting the Data Plotting Views Usin
312. um Minimal Occurrence in Single Spectra Enter in which part of the spectra a peak must occur at least in to be added to the average peak list In the case of few classes and little differences between the classes a quite high value can be chosen Note If there are more classes and the classes differ much a lower value has to be chosen e g in the case of 8 classes with total different peak patterns a value lower than 1 8 12 5 has to be chosen Aggregation Width Enter the range of clusters in ppm for peak aggregation during average peak list generation as well as during the classification in the case of the statistical test based algorithm 9 38 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Reference Part ClinProTools Menus Limit Peak Number Check this option if you want to limit the number of peaks to pick to a Maximal Peak Number according to a selected Sort Mode Maximal Peak Number Enter the maximal number of peaks to pick If more peaks have been found the first N best peaks according to the selected Sort Mode are kept You can set the number to 0 to allow pure manual peak editing Section 7 1 5 2 Sort Mode Select the sort mode according to which the N best peaks are selected Signal to Noise Sorts by signal to noise ratio Area Sorts by area Intensity Sorts by intensity In Peak Calculation define how to calculate the peaks in the individual spectra Use Areas Check this o
313. ver A peak ranking is derived from the obtained hyperplane solution The procedure is iterated until for each class a classifier class vs rest is obtained Upon the obtained SVM model the best number of peaks is determined if not manual given by a clustering in the subspace taken from the k best peaks and the best solution is stored as the final model Parameterization The parameters for the SVM are defined in the Settings Support Vector Machine dia log Section 9 1 5 2 2 The detection mode for determining the best number of peaks to be integrated in the model has to be defined you can apply the Automatic Detection 1 25 Peaks mode or specify a Number of Peaks Section 6 2 1 5 For k nearest neighbors classification Section 6 2 2 the Number of Neighbors can be set to default odd values Classification result The result of the SVM is the peak combination which is proved to separate best between the different classes 6 2 1 3 Supervised Neural Network Algorithm The Supervised Neural Network algorithm SNN is a prototype based classification algorithm If one considers a set of spectra divided into e g two classes cancer control the SNN tries to identify some characteristic spectra for each class These spectra are named prototypes and could be somehow considered as prototypical samples of that class e g the prototypical cancer patient from a proteomic point of view The determination of these prototypes is a complicated task
314. w offers the following command Command Used to Copy Copy a graphic of the focused MATLAB based window to the clipboard 9 3 1 1 Copy Command The Copy command copies a metafile graphic of the focused MATLAB based window to the clipboard This allows pasting that graphic into an appropriate application 9 3 2 View Menu The View menu of a PCA window or the Dendrogram window can offer the following commands Command Used to Mark Data Points Switch to marking data points mode applicable to PCA plots only Zoom Switch to zoom mode applicable to 2D plots only Pan Switch to pan mode applicable to 2D plots only Rotate 3D Switch to rotation mode applicable to 3D PCA plots only 9 3 2 1 Mark Data Points Command The Mark Data Points command switches to the marking data points mode This mode allows marking data points in the Scores plots with corresponding file description ga Normaho_L1 8_1SLin e g and in the Loadings plots with peak m z description e g 9 82 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Reference Part MATLAB Based Menus ests When this mode is active the first left mouse button click on a data point marks the respective point and a second click on that point removes the attached information again 9 3 2 2 Zoom Command The Zoom command switches to zoom mode which allows zooming operations zoom in out reset zooming in the 2D plots To zoom in a 2D plot move
315. w to enter the corresponding default resolution here Resolution Defaults Lists specific mass ranges for which default resolution values can be loaded into Reso lution This requires the ResolutionDefaults xml file be present in the ClinProTools folder If this file is not present the box is disabled and shows No Defaults Available To enter a default resolution select the mass range from the list In Baseline Subtraction define the parameters for the baseline subtraction filter Top Hat Baseline Select this option to perform Top Hat baseline subtraction Section 6 1 1 1 which constructs the baseline by means of morphology operators The range for the minimum and maximum search can be enlarged with the Minimal Baseline Width parameter Convex Hull Baseline Select this option to perform Convex Hull baseline subtraction Section 6 1 1 1 which constructs the baseline by fitting multiple parabolas to the spectrum The Baseline Flatness parameter influences baseline construction Minimal Baseline Width For Top Hat Baseline selected enter the minimal baseline width This parameter influences the level of details to which the baseline approaches the spectrum If this value is larger than 0 0 it tells the algorithm that the range in mass units should be at least the given fraction of the mass of the actual data point for which the baseline has to be calculated If this value is increased groups of overlapping peaks will be les
316. windows can be open at a time each representing the results of another PCA run 5 14 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH ClinProTools User Interface CIS geSpiked 0_o1 8_1SLin Figure 5 143 PCA main window showing PCA results for two loaded classes in the Scores plots top row and Loadings plots bottom row 5 2 1 2 Single Scores Plot Loadings Plot Window Each Scores plot or Loadings plot of the PCA main window can be displayed in a separate window Figure 5 14 using the corresponding command in the PCA main window s Plots menu ClinProTools User Manual Version 2 2 5 15 ClinProTools User Interface Bruker Daltonik GmbH MM Loadings 3D Figure 5 14 3D Scores plot window left and 3D Loadings plot window right 5 2 1 3 Influence Window The Influence window Figure 5 15 displays the Influence plot of the current PCA with respect to the chosen PC number You can open the window via the Influence command from the PCA main window s Plots menu i Influence Edit View ga Spiked0_018_15 ga Normand _M23_1 SLin Figure 5 15 Influence window showing an Influence plot 5 16 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH ClinProTools User Interface 5 2 1 4 Variance Window The Variance window Figure 5 16 displays the variance plot of the current PCA Section 7 5 2 3 You can open the window via the Variance command from the PCA main window s Plots
317. ws Spectra View Gel Stack View Peak Statistics View and Model List View the title bar menu bar toolbars and status bar ClinProTools Class 1 00h Class 2 02h Class 3 04h Class 4 06h Class 5 08h DAR File Edit Yiew Data Preparation Model Generation Classification Statistical Analysis Reports Compass Help Pa EOE See A E S E e e i E EE al arb u jles ClinProTools ClinProTools Test Data EDTA Run00h Sample0_G10_1SLin tid ESPE p New Show Classify 100 Load Clear All Model List Save Validate 80 Model Name Algorithm Cross Validation Recognition Capability State Modell GA 97 78 100 00 Calculated 60 Model2 SVM 81 84 100 00 Calculated Model3 SNN 62 39 100 00 Calculated Model4 ac 87 33 100 00 Calculated 40 Ee anaana a naaa aaa aaa i 1700 1800 es ClinProTools ClinProTools Test DatatEDTA RunW8h Data Files ClinProTools ClinProTools Test DataEDTA Run6h Files ClinProTools ClinProTools Test DatatEDTA RunOgh E D Data Files ClinProTools ClinProTools Test DatatEDTA Runi02h 4 D Data Files ClinProTools ClinProTools Test Data EDTA Run 00h Pk 23 1898 Da For Help press F1 X 1609 97 m z Pk 15 Y 44 52 arb u Figure 5 1 ClinProTools window with Spectra View top left Model List View top right Gel Stack View bottom left and Peak Statistics View bottom right ClinProTools User Manual Version 2 2 5 1 ClinProTools User Interface Bruke
318. x 7 14 miz Figure 9 56 Display of distance cursor in the Spectra View after selecting the Distance command 1450 1500 1550 mz X 1467 36 m z dx 45 95 m z Figure 9 57 Positioning the moveable line on the point where distance measurement should start 100 ia 1450 4500 4550 m X 1538 30 m z dx 70 94 mz Figure 9 58 Switching fixed and moveable lines and then positioning the now move able line on the point where distance measurement should end 9 74 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Reference Part ClinProTools Context Menus 1450 1500 1550 mz X 1467 36 m z dx 70 94 m z Figure 9 59 Left clicking again changes the sign of the difference value accordingly 9 2 9 11 Edit Model Name Command The Edit Model Name command is used to edit the name of a model parameter set in the model list This allows entering a name if no name was specified when adding the parameter set or changing the current name Editing the model name is possible as long as model calculation is not started The command opens the Model Name dialog Figure 9 40 to specify a model name Clicking OK enters the new model name in the model list 9 2 9 12 Edit Peak N Command The Edit Peak n command is used to change the current integration region of the selected peak Editing peaks is possible after average peak list calculation as well as after peak statistic calculation or model generation In the la
319. xactly this boundary points are removed and the SVM performs poor Different classes with pre and post treatments Currently ClinProTools considers each loaded class as independent with respect to the other classes In addition it is expected that the samples are independent to each other in the sense that e g two loaded samples do not come from the same clinical specimen Therefore ClinProTools does not completely support a scenario where one class contains samples e g cancer before clinical treatment and the second class contains samples post a clinical treatment To handle such cases a modeling of the semantic of the clinical specimens and the experimental design is necessary which will be part of a future version of ClinProTools 6 4 3 2 Small P Value Phenomenon Within ClinProTools different kinds of statistical tests are offered to the user They can be used to identify peaks which show a significant difference between the considered classes Within clinical proteomics we are in general confronted with a small set of samples and a large number of identified peaks From the identified peaks ClinPro Tools derives some characteristics such as the peak area intensity which are consid ered as features and used for the further processing These peak areas intensities are the values that are analyzed by the statistical tests as well as by the classification algo rithms The statistical tests give a p value for each peak This p
320. y run before opening the dialog ClinProTools User Manual Version 2 2 9 61 Reference Part ClinProTools Menus Bruker Daltonik GmbH Correlation Matrix I Use Kendall s Tau Algorithm Cancel The Sort Mode from the __Cancet_ Statistic Settings is Used Sort in Groups Help Correlation Level Range 0 01 1 0 MV Calculate Over All Classes Class Selection class 1 D Data Fles ClinProTools CinProTools Test Datal Spiked DatalNorme z Figure 9 48 Correlation Matrix dialog default setting Use Kendall s Tau Algorithm Check this option if the Kendall s tau b algorithm Section 6 4 2 1 should be used for correlation analysis Uncheck it to use the standard correlation algorithm Sort in Groups Check this option if the peaks should be sorted in correlation groups Correlation Level Range 0 01 1 0 Enter a correlation level as absolute correlation value for building correlation groups Recommended values are 0 7 0 95 Calculate Over All Classes Check this option if correlation should be calculated over all model generation classes Uncheck it if you want to calculate correlation over only the class selected in Class Selection Class Selection If Calculate Over All Classes is not checked select from this list the model generation class over which you want to calculate correlation OK Calculates the correlation analysis over all or the selected class and shows the results in the Correlation Matrix repor
321. y this a set of new variables so called principal compo nents PCs is generated In many cases depending on the complexity of the data set only few PCs compared to the large number of original variables contain most of the variance P value A p value is the probability that an observed effect is simply due to chance it therefore provides a measure of the strength of an association A p value does not provide any measure of the size of the effect and cannot be used in isolation to inform clinical judgment P values are affected both by the magnitude of the effect and by the size of the study from which they are derived and should therefore be interpreted with caution In particular a large p value does not always indicate that there is no association and similarly a small p value does not necessarily signify an important clinical effect QuickClassifier The QuickClassifier QC is a univariate sorting algorithm based upon classical test statistic It determines characteristics for each peak upon its statistical properties These characteristics are used to set up a model for later classification Recalibration Recalibration means the alignment of a number of spectra using the most prominent peaks A 8 ClinProTools User Manual Version 2 2 Bruker Daltonik GmbH Appendix Receiver Operating Characteristic The Receiver Operating Characteristic ROC curve gives a graphical overview about Specificity and Sensitivity of a
322. y to be important The classification and statistical testing problem becomes much easier if the number of potentially important peaks features is small and the features are not just noise To constraint C5 It is important to be careful about dependent samples Section 6 4 3 4 e g samples which are measured from the same clinical person cannot be considered to be inde pendent The same applies to multiple measurements of the same sample multiple spotting Section 6 4 3 3 This has effects on the determined features peak areas The first case has to be controlled by the user If dependent samples are used in the same class the subsequent results are more or less inaccurate The second point can be controlled by the ClinProTools multiple measurement handling If multiple measurements mm or dependent samples are considered as independent the test results can become distorted For mm which are considered to be independent we could get p values which are extremely small whereas if the mm are considered valid the p values are in correct ranges Another problem in the determined features is a large number of zero s If for a larger number of spectra at a picked peak the peak area is close to zero we are confronted ClinProTools User Manual Version 2 2 6 39 Basics Bruker Daltonik GmbH with this problem Some tests are very sensitive to a large number of close to zero values and the p values may again be unrealistic small If su

ClinProTools Manual - Vanderbilt University School of

Contents

Download Pdf Manuals

Related Search

Related Contents