Home

10.10 Setup Quant 2 Method

Contents

1. Figure 88 Setup Quant 2 Method Settings Page Marker Size Use this drop down list to set the size of the markers used in the graphical dis play of the results on the Graph page Select Preprocessing Options for Optimize Choose the method of data preprocessing from the list that is to be deployed during the optimization Select one or several methods by left clicking on the items in the list Clicking again on a selected item deselects it Note Derivatives are smoothed using the settings defined on the Parameters page Maximum Test Range You can narrow down the frequency region which will be used for an optimiza tion Bruker Optik GmbH OPUS QUANT 101 Reference Section Interactive Region Selection Alternatively to manually entering the frequency limits you can click on this button A graphical display appears from which you can select the frequency region limits interactively similar to the frequency selection on the Parameters page User defined Optimization Regions As an alternative to specifying the maximum test range you can also specify user defined frequency regions for the optimization Depending on the optimi zation type NIR General A General B you have selected on the Optimize page click either on the NIR regions max 5 or on the A B regions max 10 option button You can specify the subregions either by entering the values manually in the table or selecting the frequency
2. Validation fv lidation No 3 gt ake Calibration Methanol x Rank E DI Rec 7 Validation Report Validation Report General Information Exclude Outliers Method File Alk_d2 q2 Standards total 30 Calibration Spectra 30 Test Spectra 0 Data Block AB Compounds total 3 Frequency Regions 1 Selected Datapoints 961 Dranrnannnnina Canand Narivatiun Figure 82 Setup Quant 2 Method Report Page Spectral Residuals If you select this report type the file name the Mahalanobis distance the FProb the FValue and the residuum are listed in a table Bruker Optik GmbH OPUS QUANT 93 Reference Section Repeated Measurements If you select this report type the sample number and the corresponding standard deviation are listed in a table Print Prints the current report Exclude Outliers Select the report type Concentration Outlier and click on the Exclude Outliers button to exclude all potential outliers from the calibration set Window See section 11 3 10 8 Setup Quant 2 Method Store Method If you want to store a QUANT 2 method you have set up for future use click on the Store Method tab The method file has the extension q2 and contains all the necessary information This file can be loaded on the Load Method page Setup Quant 2 Method C OPUS 6 0 Data Extended Demodata QuantTutorial Alk_d2 x Load Method Components Spectra Pa
3. r Set subset samples on 0 4 Excluded b 0 5 amp 0 2 0 1 0 0 1 DZ DA O4 05 DE OF O8 03 Figure 68 Score Diagram Bruker Optik GmbH OPUS QUANT 79 Reference Section The scores indicate the position coordinates of the sample in a so called factor space In case the samples are close to each other or they form clusters these samples have similar spectra If you have marked the calibration spectra on the Spectra page in different col ors see section 10 3 subsection Set Color on page Graph for selected Spectra you can have the spectra displayed in the specified colors also in the score dia gram by activating the Color check box Note If you have set some calibration spectra to Excluded or Test the color dis play is deactivated Use the score diagram to display the spatial distribution of the samples for the different factors by selecting different factors in the X Axis and Y Axis drop down list On the basis of these displays now determine the factors for the auto matic selection of the subset This dialog window provides an alternative for selecting the spectra for the test data set in the best possible or excluding spectra from both the calibration and the test set Select the appropriate factor s specify the subset in for the automatic selection of the subset i e which spectra are to set to Test or Excluded and click on the Select Subset button As a result the program selects automatica
4. Loadings Special Figure 80 Setup Quant 2 Method Magnifying a Selected Area Bruker Optik GmbH OPUS QUANT 91 Reference Section Setup Quant 2 Method C OPUS SOA Data Extended Demodata QuantTuto In the Difference True plot potential outliers are depicted in red They can directly be excluded from the calibration set by double clicking on the data point with the left mouse button As a result the color will turn to black indicat ing that this spectrum has been excluded from the set Repeating the double click revokes the exclusion After excluding spectra from a calibration set repeat the validation and compare the results 10 7 Setup Quant 2 Method Report Load Method Components Spectra Parameters Validate Graph Report Store eT Optimize Settings Validation Validation No3 DI Ge C Calibration l True Prediction bo Rank 7 7 Rec 7 rename ao Prion wore 0 226 0 185 0 166 0 306 0 0663 0 0949 0 245 Exclude Outliers 0 0584 0 0169 0 00256 0 0313 0 0221 0 0683 0 0216 0 00948 0 0164 AB OSALKS 2 Figure 81 Setup Quant 2 Method Report Page The option buttons and drop down lists of the Report page are identical to the ones on the Graph page For a detailed description see section 10 6 The only difference is that the re
5. sample might consist of several spectra see figure 13 Bruker Optik GmbH OPUS QUANT 27 Validating the Model 3 The table lists the components you have entered on the Components page The validation will be performed only for the components selected with the check box This may be useful in case you are interested only in a few components to save processing time Since our calibration set con sists of only 30 spectra use all components for the validation 4 You can limit the rank to a maximum number which is specified in the Max Rank column Enter the value 0 for all components Although experience has shown that this rank might be too high for a 3 component system we recommend using it to get a feeling for this function 5 Start the validation by clicking on the Validate button This will bring up a dialog box prompting you to enter a name for the validation run with Validation No x being the default setting After clicking on the OK but ton the validation starts Set alidation Name x Please enter a name for the validation Validation No1 Cancel Figure 17 Setup Quant 2 Method Set Validation Name If you are working in a 27 CFR part 11 validated environment and your spectra are not signed an error message will occur For detailed infor mation about singing spectra and methods refer to chapter 8 6 The progress of the calculation is indicated by the status bar As you can see the algorith
6. Change Path This button allows you to change the path of your spectrum files You can change the path either for one spectrum file several or all spectrum files To do this you have to select the spectrum files in question before clicking on the Change Path button with one exception if you want to change the path for all spectrum files you need not select them To select one spectrum click on the numbered tile on the left side of the table To select several spectra separately left click on the numbered tiles while pressing the CTRL key To mark a block of spectra select the first spectra by clicking on the respective numbered tile then select the last spectrum of the block by left clicking left while pressing the Shift key 64 OPUS QUANT Bruker Optik GmbH Setup Quant 2 Method Spectra Set Data Set Clicking on the Set Data Set button brings up a dialog box figure 54 allowing to define the test set and calibration set used in a test set validation By default all spectra are assigned to the calibration set You can change the Data Set type either manually by editing the table or automatically by assigning the test set data type using this button Set Data Set x r Automatic selection of test samples First test sample fi Clear Test Spectra Clear all test spectra to use this function Block length test samples fi _Sleat Test Specta_ E Gap calibration samples fi Exchange Test lt gt Calibra
7. EE MAS42981 0 1 Aa ae ouant heri MAS 42998 0 1 Ax ap quant miston EE MAGRO Ax ae quant history EU MAS 423990 0 1 ae quant history EE MAS 42995 0 1 ae quant history EE MAS42996 0 1 Aaa QUANT berei EE MASAI Ax ae ouant berei PE MAS43010 0 1 deg QUANT berei PE MAS43011 0 1 Aa ae quant heri EE MASAI Ax ee quant berei EE MCA42992 0 1 barreari E MCA42993 0 1 Aa ae ouant bustons EE MCR40691 0 1 Aa ae ouant heri Eg MCR40692 0 1 SE MCR40704 0 1 Aa ae quant history EE MCR 40705 0 1 Aa ae quant history E MEB31233 0 1 deg QUANT berei EI MEB40612 0 1 gt Component Prediction Unit Mah Dist Limit_ Outtier Component value Densit Fett 12 61 Yo 0 068 0 29 13 71 ae quant uistory m LS 4 r Report Display full_access ows 2 Operator Default Administrator bx For Help press F1 No Active Task UE MUERE Figure 90 Quantitative Analysis Quant Report The component value density is the number of neighboring calibration spectra per component unit for example a value of 5 means that there are 5 calibration spectra per component unit e g mg l In other words this value provides information about the number calibration spectra near the predicted concentra tion value Use the component value density to distinguish between concentration regions which are well represented by ca
8. Improving the Model 8 9 The examples in table 1 show that a chemometric model can easily be improved if reasonable spectroscopic assumptions are included in the analysis However you can also have the QUANT software perform the optimization for you On the Optimize page the QUANT software auto matically checks common frequency regions in combination with sev eral data preprocessing methods The results of the optimization procedure together with the used parameters are listed in the window Note that the software yields only a list of the used parameters fre quency region and preprocessing method as well as the resulting RMSECV value and the rank the choice of the best parameters to be used for the validation is still the responsibility of the user Depending on the amount of data the optimization procedure will take a consider able amount of time In case of a test set validation it takes several min utes to perform an optimization while in the case of a cross validation the optimization may take hours Go to the Optimize page and click on the Optimize button to start the optimization procedure The progress of the optimization is indicated by a status bar The data processing methods in combination with the fre quency region as well as the RMSECV value and optimum rank are listed in the window If you click on the header of one of the first two columns you get the list sorted according to the values of this column As alr
9. Name Propanol Methanol Ethanol Unit Formatting in the Quant 2 analysis report C Default settings 5 significant digits Digits after the decimal point Se Figure 10 Setup Quant 2 Method Components Page 3 Click on the Spectra tab As you can see the table contains three col umns labeled with the component names you have entered before Now load the spectra by click on the Add Spectra button The Load File dia log box opens Navigate to the Quanttutorial folder and load all spectra 05Alkx These spectra will be added to the table Besides the component name columns there are columns labeled data set sample path and file name Note that the first column indicates the spectrum number while in the Sample column the sample numbers are listed Except for the path and file name entries all other entries can be edited by clicking on the respective table cell You can remove spectra from the table by selecting one or more press the Shift or Control key while selecting spectra and using the Delete key on your keyboard Calibration is the default setting in the column Data Set Keep this set ting because all spectra are intended for the calibration set However you need to adjust the Sample column because every sample has been measured twice as mentioned above Bruker Optik GmbH OPUS QUANT 21 Setting up a Calibration Method Setup Quant 2 Method New x Load
10. New d xj Convert 3D JCAMP File to OPUS Files Load Method Spectra Parameter Graph Store Method 3D JCAMP file C OPUS 6 0 Spektrentransfer Foss Transfer Spectra dx Browse Target path C AOPUS 6 0 SPEKTRENTRANSFER SPECTRA Browse Type of JCAMP spectra Foss B chi Bran Luebbe Others Absorbance Others Transmittance Convert Figure 114 Converting the original Spectra Indicate the directory path with the 3D JCAMP file s containing the original spectra by clicking on the Browse button right to the 3D JCAMP file field Spec ify the target path for the converted files by clicking on the corresponding Browse button Indicate the type of the original spectra by selecting the corre sponding option in the Type of JCAMP spectra drop down list Then click on the Convert button In the course of this conversion OPUS also converts auto matically the x axis unit to gu if required Note During the conversion each spectrum included in a 3D JCAMP file is stored in a separate OPUS file To load the acquired spectra click on the Spectra tab xi Convert 3D JCAMP File to OPUS Files Load Method Spectra Parameter Graph Store Method Add Master Spectra Add Slave Spectra Sample Ham 11 c Transferim 0003 0 Amylum Rotating Cup Verzamelen 0003Textt C Transfer S 0003_AHE 17 1 2003 1 Spektrum 1 0 AHE 17 1 2003 2 CATransferim 0004 0 Amylum Rotating Cup Verzamelen 0004Textt C MransteriSi 0
11. OPUS Spectroscopy Software User Manual QUANT gt lt BRUKER de 2006 BRUKER OPTIK GmbH Rudolf Plank Str 27 D 76275 Ettlingen www brukeroptics com All rights reserved No part of this publication may be reproduced or transmitted in any form or by any means including printing photocopying microfilm electronic systems etc without our prior written permission Brand names registered trade marks etc used in this manual even if not explicitly marked as such are not to be considered unprotected by trademarks law They are the property of their respec tive owner The following publication has been worked out with utmost care However Bruker Optik GmbH does not accept any liability for the correctness of the information Bruker Optik GmbH reserves the right to make changes to the products described in this manual without notice This manual is the original documentation for the OPUS spectroscopic software Table of Contents 10 About this Mantial AAA 1 Introduction to multivariate Calibration eee 3 Theoretical Background 0c ccc cece ccc rece cree eeeees 5 Chemometric Models and their Validation 26 13 3 1 Choosing Calibration Samples usura 14 3 2 Acquiring Spectra and Data Preprocessing usua 16 3 3 Validating the Model AN 18 Setting up a Calibration Method cece cece eee nes 19 Validating the Model AO 27 5 1 Performing the Validation EAE 27
12. Only the protect mode Full allows for defining a time limit for the use of a protected method i e after the time limit has run out the method in question can no longer be used The default time limit is set to one year If you want to change this default value you can either enter the desired date of expiry manually or specify it interactively using the calendar shown in figure 109 To open the calendar click on the arrow button of the drop down list Note For Identity Test methods only the protection mode Full is available If you select the Full mode the protected method can only be used for analysis purposes it can not be loaded in the Setup Quant 2 Method dia log window or in the Setup Identity Test Method dialog window If you try to do this the following OPUS message appears opus eee E Full protected method cannot be loaded in the Quant 2 Setup Figure 110 OPUS Message Note It is of crucial importance that the person who has created a Quant 2 method in our example user A keeps an unprotected version of the method file Otherwise it is not possible to have look at the settings of the method or to change them after having protected the method 122 OPUS QUANT Bruker Optik GmbH The protection modes Enlarge Method and Change Parameters are intended for the following scenario User A creates a method using cali bration spectra which he does not want to give to anybody else User B is to use the method of
13. ful to identify the optimum rank which is close to the minimum of the curve Note If you have performed a test set validation the RMSEP value is calculated whereas in case of a cross validation the RMSECV value is calculated RMSEE Rank Calibration Plots the RMSEE values versus the rank R Rank Validations and Calibration Plots the coefficient of determination R versus the rank for the test set valida tion the cross validation and the calibration Mah Distance Spec Res Validation Plots the Mahalanobis distance versus the spectral residuals The Mahalanobis distance is a measure for the similarity between the analyzed spectrum and the calibration spectra Leverage Spec Res Calibration Plots the leverage value versus the spectral residuals The leverage is a measure for the influence of a sample on the PLS model Mathematically it is the Mahalanobis distance of the single calibration samples Score Coefficients Validation and Calibration Plots the scores score of y axis versus score of x axis for the test set valida tion the cross validation and the calibration Bruker Optik GmbH OPUS QUANT 85 Reference Section Repeated Measurements Validation and Calibration Plots the deviation i e the difference between each predicted component value and the corresponding mean predicted component value of a sample versus the sample number for the test set validation the cross validation and th
14. No spectral data preprocessirn Constant offset elimination Constant offset elimination Constant offset elimination Constant offset elimination Constant offset elimination Constant offset elimination Constant offset elimination Constant offset elimination Optimize status Step 1 Cross Validation 3500 3250 3000 2750 Stop Task Abort Task Figure 87 Setup Quant 2 Method Aborting an Optimization 2500 2000 1750 1500 lekuk 100 OPUS QUANT Bruker Optik GmbH Setup Quant 2 Method Settings 10 10 Setup Quant 2 Method Settings Setup Quant 2 Method C OPUS 6 0 Data Extended Demodata QuantTute Load Method Components Spectra Parameters Validate Graph Report Store Method Optimize Settings r Graph Page Method Protection Marker Size 10 E IV Store Spectra in Quant 2 Method File Use this option only if you want to protect a method in E the mode Enlarge Method or Change Parameters r Select Preprocessing Options for Optimize IV User defined Optimization Regions e B AR EE Interactive Region Selection C AB regions max 10 eri 2 erivative Straight Line Subtraction First Derivative Vector Normalization a Maximum Test Range fi 2001 7 Interactive Region Selection M Run Optimization in Background
15. ath Filename Components _ C QUANT Ice method EisMwFett q2 Figure 39 Quant 2 Multiple File Analysis Loading a Method Quant 2 Analysis File List i xi Spectra Methods Analysis Results Analyze Print use Landscape Window Print Title I Spectral Residuals a a Ga E MAS42980 0 Pulver Av EisMwFett q2 2 was42981 0 Pulver Av EisMwFett q2 3 mas42988 0 Pulver Av EisMwFett q2 MAS42989 0 Pulver Av EisMwFett q2 5 mas42990 0 Pulver Av EisMwFett q2 Pulver Av EisMwFett q2 Pulver Av EisMwF ett q2 Pulver Av EisMwF ett q2 9 MaS43010 0 Pulver Av EisMwFett q2 MO mas43011 0 Pulver Av EisMwFett q2 MAS43025 0 Pulver Av EisMwFett q2 OO MCA42992 0 Pulver Av EisMwFett q2 MCA42993 0 Pulver Av EisMwFett q2 IA McR40691 0 Pulver Av EisMwF ett q2 15 McR40692 0 Pulver Av EisMwFett q2 OS MCR40704 0 Pulver Av EisMwFett q2 SE Sd KK KK KK KK KK K Figure 40 Quant 2 Multiple File Analysis Analysis Results 52 OPUS QUANT Bruker Optik GmbH Calibration Design A major problem in preparing a sample set used for the calibration is to avoid collinearity i e that the concentration values of the sample components must not decrease or increase proportionally to each other In case of a two compo nent system collinearity can not be avoided because if the concentration of the first component decreases consequently the concentration of the other compo
16. button if you want to exclude the spectra in the display click on Calibration gt Test if you want to assign the selected spectra to the test data set 90 OPUS QUANT Bruker Optik GmbH Setup Quant 2 Method Graph Load Method The Graph Display The validation results are plotted in a diagram Additional information are dis played in a pop up box if you position the cursor on a data point In case the RMSEP or RMSECYV values are depicted against the rank the exact values of the data points are displayed For all other display types the sample number and the sample name are stated in addition to facilitate the identification of the cor responding sample Figure 79 Pop up Box stating the exact values sample number and sample name To change the marker size click on the Settings tab see figure 88 You can magnify the displayed area by drawing a frame around the area of interest while pressing the left mouse button Clicking on the right mouse button restores the original magnification Components Spectra Parameters Validate Graph Report Store Method Optimize Settings Validation No 1 M alidation No Calibration Prediction True DI Validation Rank E DI Rec 6 RPD 45 3 Bias 0 00221 Prediction vs True Methanol Cross Validation Window Print Save Spectra M Line F Color
17. high leads to overfitting and only adds noise in fact degrades the model As a consequence there is an optimum number of factors for every system i e an optimum rank A criteria for determining the optimum rank is to look at the root mean square error of prediction RMSEP see chapter 2 for details result ing from an analysis of the test set or the cross validation If the RMSEP is depicted against the rank used in each model a minimum can be observed in this graph indicating the optimum rank 18 OPUS QUANT Bruker Optik GmbH Setting up a Calibration Method This chapter shows you how to set up a QUANT model using the data provided on the OPUS CD under the path ENHANCED DEMODATA Quanttutorial The demo data consist of spectra taken from a mixture of methanol ethanol and propanol Spectra of the pure components as well as a spectrum of a mixture containing equal parts of all alcohols are shown in figure 8 As you can see the spectra of these alcohols show considerable overlap of the peaks Four functional groups are distinguishable in the spectra COH combina tion vibrations around 4800 cm the first overtones of the CH and CH groups 6000 cm 5500cm the first overtone of the COH groups 8300 cm 6000 cm and the second overtones of the CH and CH groups 8800 cm 7800 cm Above 9000 cm there are no relevant signals Below 4000 cm the spectra show a large amount of noise and the COH vibr
18. 2 Report The QUANT software also offers the possibility to automatically analyze sev eral spectra at the same time In addition you can specify several methods used for the quantitative analysis To do this select the Quant 2 Analysis File List instead of Quantitative Analysis 2 in the OPUS Evaluate menu The following dialog window opens 50 OPUS QUANT Bruker Optik GmbH Quant 2 Analysis File List Methods Spectra Analysis Results Graph Statistics Add Spectra Load Spectra List Save Spectra List CAQUANT Quant Examples 3002_GM 0 CAQUANTiQuant Examples 3002_GM 1 CAQUANT Quant Examples CAQUANT Quant Examples 3004_GM O 3004_GM 1 15 _ C QUANTiQuant Examples 3007_GM 0 IS C QUANTiQuant Examples CA UANTIQuant Examples GZ C QUANTiQuant Examples 3007_GM 1 EEE 3008_GM 1 C QUANTiQuant Examples 3012_GM 0 CAQUANT Quant Examples CAQUANT Quant Examples 3012_GM 1 3013_GM 0 CAQUANT Quant Examples 3013_GM 1 CAQUANTQuant Examples CAQUANTIQuant Examples 3038_GM 0 3038_GM 1 CAQUANT Quant Examples 3039_GM 0 CAQUANT Quant Examples 3038_GM4 CQUANT Quant Examples CAQUANTIQuant Examples 3040_GM 0 3040_GM 1 CAQUANT Quant Examples 3041_GM 0 CAQUANTiQuant Examples CAQUANTIQuant Examples 3041_GM 1 3075_GM 0 22 C QUANTiQuant Examples 3075_GM 1 CAQUANTIQuant Examples 24 C QUAN
19. 5 13 15 111 Calibration method 19 115 Calibration model 4 15 Calibration sample 3 4 13 14 15 17 Calibration set 9 13 14 15 16 18 20 27 28 33 63 65 66 67 77 92 95 Calibration spectrum 5 6 8 11 76 78 80 85 112 Change path 64 Chemometric model 13 14 16 35 40 43 Clear selected regions 75 Clear test spectra 66 Coefficient of determination 6 29 85 86 112 Collinearity 15 16 24 53 69 Component correlation 69 Component name 21 Component table 96 Component value 5 6 8 9 10 66 67 70 71 73 77 107 131 Component value density 50 105 108 Concentration data matrix 4 111 113 Copy spectra 68 Correlation coefficient 55 56 57 70 109 111 Cross validation 9 13 18 20 27 40 63 82 85 98 D Data preprocessing 12 16 17 34 73 102 Data preprocessing method 18 39 40 60 73 97 Data set 65 70 79 Difference spectrum 8 128 Difference True 85 92 109 Display preprocessed spectra 75 E Eigenvector 4 18 Entering component values 71 Exclude outliers 95 F Factor 4 5 7 8 18 77 78 79 80 81 111 Factor analysis 78 Factor for mahalanobis distance limit 97 Factor matrix 78 Factorization 113 First derivative 17 First test sample 65 Fit True 85 FProb 9 94 108 112 Frequency range 39 40 44 60 Frequency region 12 16 17 34 35 36 38 73 74 75 97 98 99 102 103 FValue 9 10 11 94 108 112 G Gap calibration sample 6
20. 5 2 Taking a closer Look at the Eguizu 31 Improving the Model BAA 35 Generating a Report and Saving the Method 43 Performing a quantitative Analysis 2c eee eee ceeees 49 Calibration Design aia ae ee See DUE Dee E Pees ees 53 Reference Section oeus s soosse E ae a ce eee eee eae eats 59 10 1 Setup Quant 2 Method Load Method s z 59 10 2 Setup Quant 2 Method Components 0 0c e eee eee 60 10 3 1Setup Quant 2 Method i RA 62 10 4 Setup Quant 2 Method Parameters asse 72 10 5 Setup Quant 2 Method Validate 2 0 eee eee eee eee 82 10 6 Setup Quant 2 Method Graphe i024 open ede bededew taaoeeaduhe bes 84 10 7 Setup Quant 2 Method EGOI 93 10 8 Setup Quant 2 Method Store Method 2 0 0 cc eee eee eee 95 10 9 Setup Quant 2 Method Optimize 0 0 0 cece ee eens 97 10 10 Setup Quant 2 Method AAA 102 10 11 Quantitative Analysisiccas 34 ey teninieiy Sense Ree ay ed guano ee 104 10 12 Quant 2 Analysis File List Methods assa 106 10 13 Quant 2 Analysis File List Spectra 20 0 0 0c eee eee 107 10 14 Quant 2 Analysis File List Analysis Ezeitza 108 10 15 Quant 2 Analysis File List EAR 109 10 16 Quant 2 Analysis File List Statistics 00 0 0 eee eee eee 110 11 12 13 14 Abbreviations and Formulas 0c ccc ceccccccssccens 111 21 CFR part 11 Compliance sss 56 65 oicc5 o o 6 dn see aig Se
21. 6000 5000 4000 Cancel 74 OPUS QUANT Bruker Optik GmbH Setup Quant 2 Method Parameters Select Frequency Range s 3 00 Zoom A Scale all Spectra gt Shift Curve gt Crosshair gt 2 00 op Copy All 1 00 0 00 Properties 12000 11000 10000 9000 8000 7000 6000 6000 4000 Cancel Figure 63 Interactive Frequency Range Selection Clear Selected Regions You can remove an entry from the frequency regions table by selecting it and clicking on the Clear Selected Regions button or pressing the Delete button of the keyboard Display Preprocessed Spectra Before starting the validation it is possible to view the preprocessed spectra You can choose between the display of either all spectra or the spectra of every x sample or the spectra with a color flag In the latter case only the spectra highlighted by color at the Spectra page are displayed For information about how to highlight spectra by color see section 10 3 subsection Set Color on page Graph for selected Spectra Note Take into consideration that loading a large number of spectra for the dis play will take some time Therefore in cases of a large number of spectra restricting the number of preprocessed spectra to be displayed is recommendable Moreover the display of a huge number of spectra can reduce the discernability of the individual spectra Click on the Display Preprocessed Spectra button The foll
22. 7 C TransteriC AHE 18 1 2003 1 10 76 64 56 3 28 10 65 a Calibration 38 CATransteriC AHE 18 1 2003 2 10 76 64 56 3 28 10 65 a Calibration 3 C TransteriC AHE 20 1 2003 1 16 45 56 38 279 14 09 40 Calibration 10 CAransferiC AHE 20 1 2003 2 16 45 56 38 279 14 09 i Calibration D C TransteriC AHE 21 1 2003 1 13 63 58 79 3 03 14 04 12 Calibration 12 C MTransteriC AHE 21 1 2003 2 13 63 58 79 3 03 14 04 43 Calibration 13 C Transter C AHE 22 1 2003 1 13 69 59 57 3 04 13 49 44 Calibration 14 Grenet AHE 22 1 2003 2 13 69 59 57 3 04 13 49 45 Calibration 15 C Transter C ANE 03 1 2003 1 9 81 47 38 1 32 31 29 16 Calibration 16 CATransferiC ANE 03 1 2003 2 9 81 47 38 1 32 31 29 Calibration 17 CATransfer C AIB 28 12 2002 8 47 49 58 2 45 30 25 48 Calibration 18 C ATranster C AIB 28 12 2002 Sd 49 58 2 45 30 25 19 C MTransferiC AIB 02 01 2003 10 43 70 86 29 as Calibration 4 Figure 126 Spectra List 3 08 v e Now you can set up a Quant 2 method using the transferred spectra as described in the previous chapters 134 OPUS QUANT Bruker Optik GmbH Index Numerics 21 CFR part 11 28 115 3D JCAMP file 131 A Aborting an optimization 100 Add component 61 Add component columns 107 Add methods 106 Add region 74 Add spectra 63 107 Analysis result 51 Automatic selection of test samples 67 B Bias 86 93 109 110 111 Block length test sample 65 C Calibration function
23. Ext O5ALK4 2 a Calibration 5 CAOPUS 6 0 Data Ext05ALK5 1 OO Calibration 5 C AOPUS 6 0DataExt05ALK5 2 Calibration 6 CAOPUS 6 0 Data Ext 05ALK6 1 Calibration 6 CAOPUS 6 0 Data Ext 05ALK6 2 143 Calibration 7 C AOPUS 6 0 Data Ext OS5ALK7 1 Calibration 7 CAOPUS 6 0 Data Ext O5ALK7 2 Calibration 8 C OPUS 6 0 Data Ext O58LK8 1 46 Calibration 8 C AOPUS 6 0 Data Ext O5ALK8 2 Calibration 9 CAOPUS 6 0 Data Ext O54LK9 1 48 Calibration 9 CAOPUS 6 0 Data Ext O54LK9 2 49 Calibration 10 CAOPUS 6 0 Data Ext O54LK10 1 Calibration 10 CAOPUS 6 0 Data Ext 054LK10 2 Figure 50 Setup Quant 2 Method Spectra Page Spectrum List The table consists of several columns data type calibration test excluded sample number target directory file name and the names of the components that you have defined on the Components page Except for the Path and File Name entries all other entries can be edited by clicking on the respective table cell You can select either one spectrum or several spectra separately or a block of spectra To select one spectrum click on the numbered tile on the left side of the table To select several spectra separately left click on the numbered tiles while pressing the CTRL key To mark a block of spectra select the first by clicking on the respective numbered tile then select the last spectrum of the block by left clicking on it while pressing the Shift key To mark the whole tab
24. Method Parameters intensive the sign of the scores has to be reversed In case of this example data set two factors describe 99 of the information of the data set So the score values of the first two factors are sufficient to describe the differences between the spectra i e 2 values per spectrum instead of 250 data points 5 Spektren Scores Faktoren 1 Faktor1 5 216 t Faktor 2 0 216 Faktor 3 1 73E 02 Faktor 4 1 52E 02 Faktor 5 3 17E 02 2 Faktor1 5 95 Sea Faktor 2 0 103 2 Faktor 3 4 97E 04 Faktor 4 4 33E 02 Faktor 5 5 65E 03 3 Faktor1 7 731 Faktor 2 0 699 Faktor 3 3 4 Faktor 4 1 15E 02 Faktor 5 2 04E 02 5 Faktor1 7 13 Faktor2 0 441 Faktor 3 3 75E 02 Faktor 4 9 54E 03 Faktor 5 4 46E 03 Figure 67 Result of a Factorization Enter a value between 1 and 20 in the field Factors the default value 5 is acceptable for most cases and click on the Factorize button to start the PCA See figure 61 The progress of the factor analysis is shown in the status bar The result of the PCA can be displayed using two diagrams a score diagram figure 68 and a loading diagram figure 69 Click on the corresponding but ton E X Axis fi E Y Axis 2 hd I Color Exit Save Print Score 2 vs Score 1 0 5 m Automatic selection of subset Used Factors 0 E ing Subset in 7o 0 1 2 02 Select Subset
25. Method Components Spectra Parameters Validate Graph Report Store Method Optimize Settings Add Spectra Change Path Copy Spectra Window Set Sample Numbers Set Data Set Comp Correlations Print E get sanoe Le Ue rgi izei 4 Calibration 1 COPUS 6 010 054LK1 4 2 Calibration 2 CAOPUS 6 00054ALK1 2 3 Calibration 3 CAOPUS 6 010 054LK2 1 4 Calibration 4 CAOPUS 6 010 054LK2 2 5 Calibration 5 CAOPUS 6 010 054LK3 1 6 Calibration 6 CAOPUS 6 010 054LK3 2 Calibration 7 C AOPUS 6 010 054LK4 1 8 Calibration 8 CAOPUS 6 010 054LK4 2 a Calibration g CAOPUS 6 0 0054LK5 1 140 Calibration 10 CAOPUS 6 01 054LK5 2 41 Calibration 11 CAOPUS 6 010 054LK6 1 IZ Calibration IZ CANOPUS 6 0 0 O5ALK6 2 143 Calibration 13 CAOPUS 6 01 054LK7 1 44 Calibration 14 CAOPUS 6 01 05A4LK7 2 45 Calibration 15 CAOPUS 6 010 05A8LK8 1 46 Calibration 16 COPUS 6 010 05ALK8 2 Calibration 17 C AOPUS 6 010 054LK3 1 18 Calibration 18 CAOPUS 6 010 054LK9 2 149 Calibration 19 CAOPUS 6 010 054LK10 1 GO Calibration 20 CAOPUS 6 010 054LK10 2 E Figure 11 Setup Quant 2 Method Spectra Page 4 Instead of editing each row manually click on the Set Sample Numbers button to change the numbering of the samples The Set Sample Num bers window appears Indicate how many spectra per sample you have been acquired in our example enter 2 Click
26. New Eza Vv New_Diff_2 0 E New_Diff_3 0 E New_Diff_4 0 Vv New_Diff_5 0 Figure 118 Difference Spectra 128 OPUS QUANT Bruker Optik GmbH Setting up a Spectra Transfer Method 10 N 9000 8500 8000 7500 7000 6500 6000 5500 5000 4500 File Name Go to Spectra Transfer E 0003 0 E Trans_0003_AHE 17 1 2003 1 Spektrun E 0004 0 E Trans_0004_AHE 18 1 2003 1 Spektrun a v 4 gt Figure 119 Transferred Spectra In the graph of figure 119 the master spectrum and the corresponding trans ferred slave spectrum are displayed in the same color This graph allows you to see how similar the master spectrum and the corresponding transferred slave spectrum are after the transfer Now click on the Store Method tab and then on the Store Method button Setup Spectra Transfer Method New x Convert 3D JCAMP File to OPUS Files Load Method Spectra Parameter Graph Store Method Store Method Please run Calculate Transfer Model before storing the method p Transfer spectra Add Spectra Transfer Spectra Figure 120 Setup Spectra Transfer Method Store Method Bruker Optik GmbH OPUS QUANT 129 Spectra Transfer 14 2 Transferring Spectra After setting up and storing the spectra transfer method and converting the orig inal data you can transfer the original spectra to OPUS using the spectra trans fer method There are two different ways of perf
27. PCA check box you can perform a Principle Component Analysis PCA See figure 61 Similar to the PLS regression the PCA is intended to reduce the huge amount of acquired data and to describe it by as few factors as possible In contrast to the PLS regression however the component values need not to be known Note The PCA is calculated only on the basis of the calibration spectra Bruker Optik GmbH OPUS QUANT 77 Reference Section During the PCA the spectra data matrix is factorized i e it is divided into two matrixes the factor matrix loadings and the score matrix See the following figure d p B Spectra Data Matrix Oo d GO n n Data Matrix n spectra with p data points Scores d scores for each spectrum d lt n Factors d factors with p data points d lt n Figure 66 Factorization of the Spectra Data Matrix During the factor analysis a set of spectra is transformed in factors loadings and the corresponding scores The factor analysis is a variance analysis i e the differences between the spectra are determined and reproduced in form of fac tors The first factor describes the as big as possible part of the whole variance the second factor the as big as possible part of the remaining variance and so on The part of whole variance that the following factors represent is becoming smaller and smaller until they represent only noise The factors are orthogonal i e they are independent so that a p
28. Protein 4 CAQUANT Quant Examples Sunflower Seeds grinded Protein 4 q2 Protein 5__ CAQUANT Quant Examples Sunflower Seeds grinded Protein 5 42 Protein 6 CAQUANT Quant Examples Sunflower Seeds Oil 1 q2 Oil CA QUANT Quant Examples Sunflower Seeds Oil 2 q2 Oil EES QUANT Quant Examples Sunflower Seeds Oil 3 q2 E oil OG C QUANT Quant Examples Sunflower Seeds Oil 4 q2 L Oil Figure 91 Simultaneous Evaluation of several Quant 2 Methods Methods Page Add Methods Click on this button to specify the QUANT methods you want to use for the analysis Load one or several methods from the load file dialog box The selected methods are listed in form of a table stating the file name path and the components of the methods Load Method List Click on this button to load a saved method list containing several method files Save Method List Click on this button to save the loaded method in a method list for future use The method list will be stored in a file with the extension q2 Clear Click on this button to delete the complete method list table Bruker Optik GmbH OPUS QUANT 105 Reference Section 10 13 Quant 2 Analysis File List Spectra Quant 2 Analysis File List b xj Methods Spectra Analysis Resuits Graph Statistics Load Specta List Save Spectra List Add Component Columns File Name CAQUANTIQuant Examples 3002_GM 0 2 CAQUANT Quant Examples 3
29. WSS eis esis 115 12 1 Siging Specta EE 115 12 2 Signing Methodsa reest ied ebakidurei 116 Method Protection era ae iee e E b E ebate 119 Spectra Transter sje tise koe sy oie E hs oe AA E See EE baia 125 14 1 Setting up a Spectra Transfer MgO 125 14 2 Transferring Spectra 14 3 Setting up a Quant 2 Method using transferred Spectra 131 About this Manual This manual is divided into four parts The first part chapter 1 to 3 explains the theory of the OPUS QUANT software Moreover it contains information about multivariate calibration the different kinds of validation and data preprocessing methods The second part chapter 4 to 9 is a tutorial that provides a step by step intro duction to the QUANT analysis using the example data provided on the OPUS CD You will find these data under ENHANCED_DEMODATA Quanttuto rial In this way you can reproduce the examples on your own computer while working through the corresponding chapters of the manual The third part chapter 10 to 11 servers primarily as a reference you can consult if you have questions about a function or a particular problem with operating the OPUS QUANT software Chapter 10 describes all QUANT functions in a systematic manner Chapter 11 provides definitions and mathematical formulae of the statistical parameters that are relevant to the assessment of a QUANT method The fourth part of the manual describes how to sign spectra and methods in o
30. a calibration function b Y X 5 2 1 Y Spectrum 1 Y Spectrum 2 2 2 Y Spectrum 3 The vector Y consists of the component values of a single component as deter mined by the reference measurements The row vectors of the matrix X are formed from the calibration spectra The aim is to determine the vector b When b is known the prediction of unknown values for Y can be done The solution of the above system of equations is given by be GE Ry X FY 2 3 The PLS Method During PLS regression the matrices X are reduced to only a few factors The difficulty is the inversion of the matrix AIN The PLS method involves the cal culation of a restricted inverse instead of the complete PLS requires the matrix X is bi diagonalized X UBV 2 4 The matrices U and V are orthonormal and B is of bi diagonal form This can also be expressed as X TV 2 5 The elements of the matrix T are known as scores and the PLS vectors are sometimes called loadings A starting vector v4 for the PLS analysis is chosen T e a 2 6 v e Bruker Optik GmbH OPUS QUANT 5 Theoretical Background The first PLS vector shows the correlations between the component values and the spectral intensities of the calibration spectra The PLS analysis can be termi nated if the component values Y are reproduced in a consistent way with the help of the vector b regression The number of PLS vectors used is defined
31. a wider concen tration range than you intend to analyze later This helps to create a more stable model for analysis This becomes increasingly important if you expect outliers with con centrations that largely deviate from your desired values as this may be the case in quality control The calibration samples should be spaced homogeneously across the concentration range Do not include samples with concentrations well apart from the concentration field the majority of your samples span In case you need to extend the concentration range include a larger number of samples so that the resulting range still retains the sample density Do not try to correct external fluctuations as this will be mirrored as concentration fluctuations in your samples These fluctuations will be recognized as such by QUANT and accounted for in the calibra tion function This will yield a more robust model Keep in mind that an extensive sample preconditioning of the calibration samples will have to be repeated later for every sample to be analyzed Never try to account for deviations in the calibration set you can not correct for the samples you want to analyze Rather increase the number of sam ples included in your calibration set If your process conditions change later there is no need to repeat the calibration because the perturbations will be filtered by the PLS 1 algorithm If your concentration range expands in the future simply add a suffi
32. components used for the method as well as the frequency region the number of data points employed for the validation and the used data preprocessing method 10 2 Setup Quant 2 Method Components Setup Quant 2 Method C OPUS 6 0 Data Extended Demodata QuantTutorial Alk_ xj Load Method Components Spectra Parameters Validate Graph Report Store Method Optimize Settings Add Component Name Unit M ethanol m Formatting in the Quant 2 analysis report E Default settings 5 significant digits So Propanol lt Digits after the decimal point de Figure 49 Setup Quant 2 Method Components Page 60 OPUS QUANT Bruker Optik GmbH Setup Quant 2 Method Components Add Component Use the Add Component button to add a new entry to the components list The default name displayed in the Name field is Comp x Name The name of the component to be added or an selected entry of the list can be changed in the Name field Unit This field serves to specify the unit with mg being the default setting for the component values Removing Components To remove an entry from the list select this entry using the mouse and press the Delete key on your keyboard Arranging Entries The order of the components in the list can be changed by dragging the items with the mouse Formatting in the Quant 2 Analysis Report The formatting of the prediction
33. corresponding sample spectrum lt File name of first sample spec trum gt _AV 68 OPUS QUANT Bruker Optik GmbH Setup Quant 2 Method Spectra Note If there are samples with only one acquired spectrum no mean spectrum is calculated In this case the original spectrum is copied in new directory keeping the same file name Note The creation of a new mean spectra based Quant method has no effects on the original method and spectra They are still available for further use Copy Spectra and current Method File Store new method x r Copy spectra and current method file This function copies all spectra of the spectra list and the current method file to the destination path e g to archive the data Select the destination path or type it in then click on Start Copy Select Path E ee r Store new method based on the mean sample spectra This function calculates the mean sample spectra and stores them in a subdirectory Then a new Quant 2 method file is created Store New Method Figure 57 Copy Spectra and Method Comp Correlations Sometimes there is an unwanted collinear correlation between the sample com ponents i e the concentrations of the components increase or decrease in the same way over the sample set Collinearity must be avoided as otherwise no independent calibration can be established To check the correlation click on the Comp Correlations button A graph ap
34. for the calibration design Enter the wanted sum of component values for one sample In many cases this is 100 percent 56 OPUS QUANT Bruker Optik GmbH After clicking on the Search Component Value button switch to the Graph page As you can see there is a high correlation between Component 2 and Component 3 i e the concentration values are not distributed evenly over the complete range but have the shape of a line See figure 47 Calibration Design E Setup Table Graph 10 9417 Comp 3 Comp 2 A High Correlation Print Comp 3 vs Comp 2 ES 0 9417 Figure 47 Quant 2 Calibration Design Graphical Display of the Distribution The correlation coefficient is well above 0 7 and the High Correlation warning is displayed The chosen minimum and maximum concentration values see figure 46 do not allow to find concentration values without high correlation Therefore you should enter different minimum and maximum concentration values for one of these components 2 or 3 and repeat the calibration design Bruker Optik GmbH OPUS QUANT 57 Calibration Design 58 OPUS QUANT Bruker Optik GmbH Setup Quant 2 Method Load Method 1 0 Reference Section 10 1 Setup Quant 2 Method Load Method Setup Quant 2 Method New j x Load Method Components Spectra Parameters Validate Graph Report Store Me
35. in individual Regions 72 OPUS QUANT Bruker Optik GmbH Setup Quant 2 Method Parameters The usage of this option implies a two step data preprocessing procedure 1 Step Data preprocessing on the basis of your own defined prepro cessing sequence i e your own defined combination of preprocessing method s and frequency region s The individual preprocessing sequence is defined in the dialog window shown in figure 62 Note Defining an individual preprocessing sequence requires experiences in set ting up a calibration method as OPUS does not check whether the self defined preprocessing sequence is appropriate For detailed information about the differ ent data processing methods refer to chapter 1 2 Step Data preprocessing on the basis of the frequency region s you have defined for the calibration See the dialog window shown in figure 61 For this step you can but you need not necessarily select a data preprocessing method Note The spectra preprocessed in individual regions step 1 are the basis for the subsequent data preprocessing in calibration regions step 2 This two step procedure allows an individual preprocessing of the original data in the run up to the actual calibration In contrast to the fixed data preprocessing options in the calibration regions an additional data preprocessing in individual regions provides a greater flexibility in combining data preprocessing methods and frequency regions e
36. in the QUANT program by the size of the rank Optimum PLS rank can be calculated only if the number of cal ibration spectra is sufficiently high e g one component and 20 calibration spectra The PLS regression has the advantage that the PLS factors are arranged in correct sequence according to their relevance to predict the com ponent values The first factor explains the most drastic changes of the spec trum The residual Res is the difference between the true and the fitted value Thus the sum of squared errors SSE is the quadratic summation of these values SSE gt Res 2 7 The root mean square error of estimation RMSEE is calculated from this sum with M being the number of standards and R the rank B 1 RMSEE lace 2 8 The coefficient of determination R gives the percentage of variance present in the true component values which is reproduced in the regression R approaches 100 as the fitted concentration values approach the true values R ei x 100 2 9 KAE R can be negative This is true in some cases for low ranks when the residu als are larger than the variance in the true values y The sum of residuals SSE decreases with increasing rank so R approaches a limiting value of 100 An important measure is the Leverage value h h diag UU 2 10 The A values are a measure of the influence a spectrum has on the PLS model for a particular component A large value can ar
37. is then tested with the test set This procedure is called test set validation The distribution of the concentration values should be similar for both sets A test set validation requires less computational time than a cross val idation If only a limited number of samples is available use a cross validation see above To perform a good cross validation the number of spectra per sample should be equal for all calibration standards Important Repetitive spectra of one sample must be assigned as one sample A matrix is formed from the spectral data of the calibration set The matrix will be transformed by the PLS 1 algorithm into a result matrix consisting of eigen vectors factors only as mentioned above These factors are sorted in decreas ing order according to their contribution to the spectral features Factors which present a large contribution to the spectrum are found in the top rows of the matrix while factors listed towards the bottom rows mainly reflect spectral noise and fluctuations Thus not all factors are needed to explain the spectral features of the components the contributions representing noise can be omit ted The quality of the chemometric model now depends on the choice of the correct number of factors needed this is also called the rank of the model Choosing a too small rank results in underfitting so that not all features can be explained by the model On the other hand including too many factors rank too
38. list there is an empty row which can be used to paste data from the clipboard to enlarge the spectra list The fields Data Set and Sample will be set automatically when you return to the Spectra page The fields Path Filename and the component values can be edited By default Data Set is set to Calibration and the Sample numbering is consecutive for the added spectra 70 OPUS QUANT Bruker Optik GmbH Setup Quant 2 Method Spectra Entering Component Values Component values can either be entered manually or pasted into the spectrum list from the Windows clipboard If you paste values into the table using the shortcut Ctrl V place the cursor in the table cell in which the first of the val ues is to be pasted To enter decimal numbers you can use both comma and dot The QUANT soft ware enables you to duplicate identical entries to duplicate one row select the row in question by clicking on the cell and expanding the selection frame to the whole row while keeping the left mouse button pressed There is a small black square on the lower right corner of the frame Positioning the cursor on this square changes the pointer shape to a cross Now left click on the square while expanding the frame to the next row Upon releasing the left mouse button the values are copied in the next row Figure 60 Copying Table Entries Removing Spectra You can remove one or more spectra from the table by selecting the
39. s for Quantitative Analysis 2 field of the dialog window Load the Quant 2 method you want to use by clicking on the Load Quant 2 Method button Note that a previously loaded method is automatically loaded If you want to use another method click on the Load Quant 2 Method button and select another one Then click on the Analyze button to start the analysis Quantitative Analysis 2 x Select File s m File s for Quantitative Analysis 2 m Loaded Quantitative Analysis 2 Method C QUANTSIce method EisMwFett q2 Load Quant 2 Method Analyze Cancel Help Figure 36 Quant 2 Analysis Bruker Optik GmbH OPUS QUANT 49 Performing a quantitative Analysis The result of the quantitative analysis is appended to the respective file in form of a QUANT report block Clicking on this report block automatically opens a report window Select PLS Analysis Report to display the analysis results The upper subwindow displays the method file and information about the method used In the lower subwindow the predicted concentration value the unit the Mahalanobis Distance Mah Dist the threshold value Limit for the outlier identification and the Component Value Density of each component are listed For detailed information refer to chapter 2 ba Report Display full_access ows 2 Operator Default Administrator ZO Eile Edit View Window Measure Manipulate Evaluate Display Print Mac
40. second and third plot type requires the true component values i e these values have to be entered in the spectra list before starting the analysis In addition offset slope and correlation coefficient of the regression line as well as bias RPD and RMSEP for the predictions of the independent samples are given For detailed information about these statistical values refer to chapter 12 Save Click on this button to save the currently displayed plot as bitmap file Print Click on this button to print out the currently displayed plot 108 OPUS QUANT Bruker Optik GmbH Quant 2 Analysis File List Statistics 10 16 Quant 2 Analysis File List Statistics This page provides an overview of the calculated statistical values RMSEP bias and RPD as well as offset and slope of the regression line for all methods and components allowing a comparison of the different Quant 2 methods In addition the number of spectra used for the statistical calculation is also dis played in the table The capability of a method is identifiable by the RMSEP value the bias value and the RPD value The most capable Quant 2 method is the one with the lowest RMSEP value and the highest RPD value Moreover the bias value should be as Close as possible to zero See figure 95 The most capable method The most capable method for the component Protein for the component Oil Quant 2 Analysis File List Methods Spectra Analysis R
41. set The assignment is done automatically accord ing to the selected value in percentage for the Test Samples To use this func tion the following preconditions have to be fulfilled e No spectrum in the spectrum table has to assigned to the data set type Test e The sample set has to comprise at least 16 samples e There have to be at least four different component values for each component If there are already some test spectra you have to assign them to the calibration set by clicking on the Clear Test Spectra button Otherwise you can not use this function If there are not enough samples a minimum of 16 samples or enough different component values a minimum of four different component values for a component you can also not use the Automatic Selection of Test Samples func tion If the above mentioned preconditions are fulfilled you can select a value per centage value from Test Samples drop down list e g the option 54 means that 54 of the sample set is assigned by the software to the test set Click on the Select Test Samples to effectuate the automatic selection of the test samples Set Selected Spectra on This function facilitates the assignment of the spectra to certain data set type calibration test or excluded as all selected spectra are assigned to the specified data set type at once Bruker Optik GmbH OPUS QUANT 67 Reference Section You can use this function only if you have selected one or more
42. spectra in the spectrum table beforehand Otherwise this function is grayed i e it is deacti vated To select one spectrum click on the numbered tile on the left side of the table To select several spectra separately left click on the numbered tiles while pressing the CTRL key To mark a block of spectra select the first by clicking on the respective numbered tile then select the last spectrum of the block by left clicking left while pressing the Shift key Set Color on page Graph for selected Spectra This function allows you to highlight one or more spectra by one or more differ ent color s You can either choose an option of the drop down list blue magenta orange cyan or gray or invoke a color palette by clicking on the Color button To implement the color setting in the spectra table click on the Set Color button To undo the color setting click on the Clear Color Setting button You can use this function only if you have selected one or more spectra in the spectrum table beforehand Otherwise this function is grayed i e it is deacti vated To select one spectrum click on the numbered tile on the left side of the table To select several spectra separately left click on the numbered tiles while pressing the CTRL key To mark a block of spectra select the first by clicking on the respective numbered tile then select the last spectrum of the block by left clicking left while pressing the Shift key The color setting has
43. the file names of the spectra as well as the component values of the indi vidual components Open this text file using a normal text editor Copy the content of the text file except for the first row into the clipboard See figure 123 Bruker Optik GmbH OPUS QUANT 131 Spectra Transfer P calset Bran Luebbe_info txt Editor EISE Datei Bearbeiten Format Ansicht trum trum trum trum trum trum trum trum trum DTD DDD DDD trum trum trum trum LTZ rum 1 fo Se Tete Te hey A rum rum rum rum rum rum rum HEART Figure 123 Text File e Click in the Setup Quant 2 Method window on the Spectra tab and then on the Window button As a result of this the QUANT setup assistant is embedded in an OPUS window Among other columns the table includes also columns labeled with the component names you have entered before See figure 124 ba Quant Report full_access ows 2 Operator Default Administrator g x if Eile Edit Heu Window Measure Manipulate Evaluate Display Print Macro Validation Setup Help 8X PERITO ARNA RSH Ses eee KS Sic Ee OPUS Browser E Display full_access ows 1 Operator Default Administrator fa Quant Report full access ows 2 Operator Default Administrator Go back to A Spectra Graph Report BE EEE EGZ Figure 124 QUANT Setup Assistant e Position the cursor in the empty cell of the Path column and paste the content of the clip
44. to assess and compare several Quant methods with each other For detailed information about these parameters refer to chapter 11 86 OPUS QUANT Bruker Optik GmbH Setup Quant 2 Method Graph Window See chapter 10 3 section Window Print Prints out the current graph with the respective parameters e g Rank R RMSECY etc Save Allows you to save the graphic result as a bitmap A Save File dialog box opens Enter a file name and specify the target directory Spectra Clicking on this button opens a display window that is already described in chapter 10 4 section Display Preprocessed Spectra Loadings Clicking on the Loadings button opens the following window 12000 11000 10000 9000 8000 7000 6000 5000 4000 Alk_d2_Methanol_RegrCoeff 0 Go to Setup Quant KE Ak d2_Methanol_PLS_V1 0 m Alk_d2_Methanol_PLS_ v2 0 HO f __ Alk_d2_Methanol_PLS_V3 0 E AlkK_d2_Methanol_PLS V4 0 E Alk_d2_Methanol_PLS_V5 0 zi Interactive Region Selection Figure 73 Display the Regression Coefficient and the PLS Vectors Bruker Optik GmbH OPUS QUANT 87 Reference Section The graph in the upper part of the window shows the vector of the regression coefficient b red and the PLS vectors up to the rank you have selected on the previous page The b vector is a graphical display of the calibration function It shows the wavenumbers at which relevant informati
45. 002_GM 1 16 579 51 87 IZ C QUANT Quant Examples 3004_GM 0 13 611 53 44 I C QUANT Quant Examples 3004_GM 1 13 611 53 14 GOGUAN Quant Examples 3007_GM 0 14 334 53 82 6 CAQUANT Quant Examples 3007_GM 1 14 334 53 82 CAQUANT Quant Examples 3008_GM 0 12 865 55 51 CAQUANT Quant Examples 3008_GM 1 12865 5551 I C QUANT Quant Examples 3012_GM 0 45 137 53 02 CAQUANT Quant Examples 3012_GM 1 15 137 53 02 C AQUANT Quant Examples 3013_GM 0 15 081 49 56 CAQUANT Quant Examples 3013_GM 1 15 081 149 56 CAQUANTQuant Examples 3038_GM 0 14 548 50 51 CAQUANTIQuant Exemples 3038_GM 1 14 548 50 51 CAQUANTIQuant Examples 3039_GM 0 16 384 499 CAQUANT Quant Examples 3039_GM A 16 384 49 9 CAQUANTIQuant Exemples 3040_GM 0 17 651 44 33 CAQUANT Quant Examples 3040_GM 1 17 651 44 33 CAQUANT Quant Exemples 3041_GM 0 j15691 48 36 Protein CAQUANT Quant Examples 3041_GM 1 15 691 48 36 CAQUANT Quant Examples 3075_GM 0 14117 51 49 CAQUANTIQuant Examplest 3075_GM 1 14117 51 49 GOGUAN Quant Examples 3076_GM 0 17 698 46 66 CAQUANTIQuant Examplest 3076_GM 1 17 698 46 66 GOGUAN Quant Examples 3077_GM O 47 14 5018 CAQUANT Quant Examples 3077_GM 1 17 14 50 18 Figure 92 Simultaneous Evaluation of several Quant 2 Methods Spectra Page Add Spectra Clicking on this button opens a standard load file dialog box from which you can select the spectra you want to analyze The loaded spectra are lis
46. 004_AHE 18 1 2003 1 Spektrum 2 _ 0 AHE 18 1 2003 3 C ATransferim 0008 0 Amylum Rotating Cup Verzamelen 0008Textt 3__ C Transfer S 0008_ANE 03 1 2003 1 Spektrum 3 _ 0 ANE 03 1 2003 14 C ATransferim 0009 0 Amylum Rotating Cup Verzamelen 0009Text1 C Transfer S 0009_AIB 28 12 2002 1 Spektrum 4 _ 0 AIB 28 12 2002 ZZ C ATransferi 0010 0 run Rotating Cup Verzamelen 001 0Text 5__ C Transfer S 0010_AIB 02 01 2003 1 Spektrum 5 _0 AIB 02 01 2002 E C ATransferim 0011 0 Amylum Rotating Cup Verzamelen 0011 Textt E C Transfer S 0011_AIB 04 01 2003 1 Spektrum 6 _ 0 AIB 04 01 2002 CATranster 0013 0 Amylum Rotating Cup Verzamelen 001 3Textt C Transter S 0013_ANE 14 01 2003 1 Spektrum 7 _ 0 ANE 14 01 200 C TransferiM 0026 0 Amylum Rotating Cup Verzamelen 0026Text1 E C Transfer s 0026_ABU 11 01 2003 1 Spektrum 8 _ 0 ABU 111017200 a C ATransferim 0028 0 _ Amylum Rotating Cup Verzamelen 0028Textt GZ C Transterisi 0028_ABU 01 02 2003 1 Spektrum 9 _ 0 ABU 01 02 200 C ATransferiM 0033 0 Amylum Rotating Cup Verzamelen 0033Textt 10 _ C Transfer S 0033_ASL 8 01 2003 1 Spektrum 10 _ 0 ASL 8 01 2003 Figure 115 Setup Spectra Transfer Method Spectra 126 OPUS QUANT Bruker Optik GmbH Setting up a Spectra Transfer Method The spectra acquired using the Bruker spectrometer are the master spectra and the spectra measured usin
47. 08 opens Select the corresponding method type Quant 2 Method q2 or Identity Test Method faa and load the method file from the appropriate directory e g USER_B 00 00 AD 07 AB 11 by clicking on the Load Method button Methods Add Signature Show History C OPUS 6 0 Data Extended Demodata QuantTutori Signature History Load Method Quant 2 Method q2 b Add Signature Protect Method Print Signature The method file has no signature block Figure 108 Methods Add Signature Show History Dialog Box Bruker Optik GmbH OPUS QUANT 121 Method Protection 5 Click on the Protect Method button The dialog box shown in figure 109 opens Enter the MAC ID of the spectrometer of user B in the corre sponding entry field with the blanks as shown in the figure 109 Prepare Method Protection x For safety reasons always make a copy before protecting a method Protect mode Full b Time limit 09 03 2007 e Change Parameters Marz 2007 C Enlarge Method Di Mi Do Fr Sa 6 27 28 1 2 3 6 7 8 EN 10 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 3 1 Call Add signature to write the protect settingsto lt 4 5 b Z Heute 09 03 2006 Mag 0000AD 074811 eg 00004 Set Cancel Help Figure 109 Prepare Method Protection There are three different protection modes e Full e Enlarge Method e Change Parameters
48. 1 2003 2 13 59 59 57 3 04 13 48 03 CATransferic ANE 03 1 2003 1 9 81 4738 132 31 28 0 45 CATransferiC ANE 03 1 2003 2 9 81 4738 EJ 31 28 0 45 Gere AIB 28 12 2002 18 47 49 58 2 45 30 25 13 Gere AIB 28 12 2002 18 47 49 58 2 45 30 25 13 Gere AIB 02 01 2003 10 43 70 86 23 3 08 18 Gere AIB 02 01 2003 10 43 70 86 23 3 08 18 Gezi AIB 04 01 2003 ZEA 49 54 2 23 33 23 117 C ATransferiC AIB 04 01 2003 ZEA 49 54 2 23 33 23 117 Gere ANE 9 01 2003 1 8 41 6181 156 15 07 05 Garzi ANE 9 01 2003 2 8 41 61 81 156 15 07 05 CAMTransferic ANE 14 01 2003 9 03 62 54 151 1452 0 85 CAMTransteric ANE 14 01 2003 9 03 62 54 151 1452 0 85 CAMTranster C ANE 15 01 2003 8 77 60 74 152 1579 083 CAMTransteric ANE 15 01 2003 8 77 60 74 152 1579 083 CAMTransferiC ANE 16 01 2003 9 08 604 153 1582 075 CATransferiC ANE 16 01 2003 9 08 604 153 1582 075 CAMTransfer c AET ZOZO SEE 61 83 154 16 03 031 CAMTransfer C ANE 17 01 2003 9 08 61 83 154 16 03 031 CATransteriC ANE 21 01 2003 6 23 63 41 157 16 37 KO CATranster C ANE 21 01 2003 6 23 63 41 157 16 37 KO CATransteric ANI 26 12 2002 15 58 61 84 186 931 12 CATransteric ANI 26 12 2002 15 58 61 84 186 931 12 ei d A Quant Report full_access ows 2 Operator Default Administrator l dbx No Active Task r lino fer FE e Click in the QUANT setup assistant window on the Spectra button As a result of this the Spectra page of the Setup Quant 2 Method window appears In the spectra table the path
49. 100 but can also be user defined These optimized concentration values can be used to set up the calibration The table can be printed by clicking on the Print button or copied to the clipboard Ctrl C and pasted Ctrl V into other applications On the Graph page a graph showing the concentration distribution of the com ponent pairs and the corresponding correlation coefficient are displayed Select the wanted component pair from the drop down list Bruker Optik GmbH OPUS QUANT 55 Calibration Design Calibration Design E Setup Table Graph Comp 2 vs Comp 1 SS 0 0213 15226335445555665 7758864995 Figure 44 Quant 2 Calibration Design Graphical Display of the Distribution If the correlation between two components exceeds the threshold correlation coefficient larger than 0 7 a warning message figure 45 will be displayed ZN High Correlation Figure 45 High Correlation Warning To generate an example for a data set with a high correlation enter the follow ing minimum and maximum concentration values Number of Components 3 E Sum of component values 100 b Uniform distribution of concentration values Figure 46 Quant 2 Calibration Design Setup Number of samples range 15 100 D Search Component Values Instruction Select the number of components used
50. 4 914 05A1k12 0 1 25 425 0 74 575 05A1k13 0 1 33 394 66 606 0 05A1k14 0 1 0 65 944 34 055 05A1k15 0 1 33 104 33 641 33 254 7 Avoid collinearity i e ensure that the concentrations of the components do not increase or decrease in the same way over the sample set Other wise no independent calibration can be established To check the corre lation click on the Comp Correlations button A window appears showing the concentration distribution of the samples for each compo nent pair 24 OPUS QUANT Bruker Optik GmbH EE 0 2499 Ethanol Methanol x Ethanol vs Methanol r 0 2499 SEA 0 0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 105 Figure 15 Setup Quant 2 Method Collinearity Check In our sample set the concentration values are evenly spread and no col linearity can be observed The R value squared correlation coefficient is well below 0 7 the threshold for correlation If this value is above 0 7 the following warning will be displayed A High Correlation In this case a review of the prepared samples will be necessary The OPUS function Calibration Design helps you to find the optimal con centration values for a set of samples beforehand See chapter 7 8 To generate a calibration model using the entered data click on the Vali date tab Bruker Optik GmbH OPUS QU
51. 5 I Interactive region selection 35 74 103 L Lambert Beers law 3 Leave excluded spectra 66 Leverage value 6 7 Leverage Spec Res 85 Linear offset subtraction 17 Load existing validation results 60 Load method 59 Load method list 106 Load Quant 2 method 104 Load spectra list 107 Loading diagram 79 Loadings 5 78 81 87 111 M MAC ID 119 120 122 124 Mah Distance Spec Res 85 Mahalanobis distance 8 50 85 94 105 108 112 Marker size 102 Master spectrum 127 129 Mean centering 73 Mean spectrum 17 68 69 73 77 78 Method protection 103 119 Min max normalization 17 Multicomponent system 3 4 13 15 Multiplicative scatter correction 17 Multivariate calibration 3 4 O Offset 93 109 110 112 Optimization 40 97 98 102 Optimization results 100 Outlier 3 7 8 9 11 12 15 29 50 63 77 92 94 95 105 108 112 113 Overfitting 4 18 P PCA 77 78 79 PLS algorithm 4 15 16 18 111 PLS factor 6 PLS method 5 PLS model 6 12 73 85 PLS rank 10 PLS regression 3 4 5 6 16 35 77 PLS vector 5 6 8 9 88 112 113 Prediction 5 9 10 13 18 31 32 33 86 98 108 112 113 114 Prediction error 10 11 Prediction value 20 21 61 Prediction Sample number 109 Prediction True 31 85 86 109 Preprocessing in calibration regions 73 Preprocessing in individual regions 72 Preprocessing sequence 73 PRESS 10 112 Principal component 4 Principle component analy
52. ANT 25 Setting up a Calibration Method 26 OPUS QUANT Bruker Optik GmbH Performing the Validation Validating the Model 5 1 Performing the Validation Proceed with the example of chapter 4 Click on the Validate tab Load Method Components Spectra Parameters Validate Graph Report Store Method Optimize Settings m Validation parameters Component max Rank Use Methanol 10 E 2 Ethanol OIS Gu Propanol Cross Validation F No of samples leaving out fi Setup Quant 2 Method C OPUS 6 0 Data Extended Demodata QuantTutor al Al x Calculation status Validate Figure 16 Setup Quant 2 Method Validate Page 1 The window comprises two group fields the Validation Parameters and the Calculation Status Select the method used for the validation from the drop down list You can choose between Cross Validation and Test Set Validation with cross validation being the default setting For our example use this validation type If you chose Test Set Validation instead you have to indicate on the Spectra page which samples form the calibration set and which the test set 2 As explained in chapter 3 a number of spectra has to be excluded from the calibration set that will serve as internal test samples Specify the number of samples to exclude per cycle in the No of samples leaving out field For our example use the default setting a
53. Data Ext O54LK8 2 Calibration 9 CANOPUS 6 0 Data Ext 05ALK91 148 Calibration 9 CAOPUS 6 0 Data Ext O54LK9 2 149 Calibration 10 CAOPUS 6 0 Data Ext 054LK10 41 GO Calibration 10 CAOPUS 6 0 Data Ext O54LK10 2 gt Figure 13 Setup Quant 2 Method Spectra Page 5 Enter the concentration values for each sample The alcohol concentra tion values for all 15 samples are listed in Table 1 To facilitate this task you can duplicate identical entries by clicking on a cell There will be a small black square on the lower right corner of the cell Position the cur sor on this square As a result the pointer shape changes to a cross Now press the left mouse button while expanding the frame to the next cell in order to copy the content of the cell In this way you can also copy a row Figure 14 Copying Table Entries Bruker Optik GmbH OPUS QUANT 23 Setting up a Calibration Method Table 1 Component Concentrations of Example Files 6 To print the data you have entered click on the Print button File Name Methanol Ethanol Propanol 05A1k1 0 1 0 0 100 OSAIk2 0 1 100 0 0 05A1k3 0 1 0 100 0 05A1k4 0 1 33 364 33 278 33 356 05A1k5 0 1 49 666 25 435 24 899 05A1k6 0 1 24 942 24 982 50 078 05A1k7 0 1 26 392 48 95 24 658 OSAIk8 0 1 50 017 0 49 983 OSAIk9 0 1 66 648 33 352 0 05A1k10 0 1 0 33 392 66 606 05A1k11 0 1 75 086 0 2
54. ER P Quick Print Display full_access ows 1 Operator Default Admini 2 Operator Default 4 New Layout E Open Layout n Report bre Print Preview Gener Brink p Print Setup Method Fe Ak OO ZEZ x Standards total 30 Calibration Spectra 30 I Test Spectra 0 Data Block AB N Components total 3 ENa Frequency Regions 1 Selected Datapoints 2076 N Mean Centering Yes Rreprocessing to 12001 7 3999 3 Component Range kr Compound Unit Validation Type 4 4 Nency Regions ethanol 0 100 IN Ma of samples leaving out gt Quant Report full_access ows 2 Operator Default Administrator bx la x ZZ No spectral data preprocessing Soss Validation Print the active document F num RF NoNictive Task Setup Quant Go back to Figure 34 Setup Quant 2 Method Printing the Report cee Spectra Graph Report 46 OPUS QUANT Bruker Optik GmbH 5 6 7 Finally save the Quant 2 method you have created so far by switching to the Store Method page This page displays a summary of the relevant information about the selected validation You can store this information including the validation results by activating the respective check box and then clicking on the Store Method button Thereupon a method file with the extension q2 is generated that will be used
55. Sample After you have entered the values for First Test Sample Block Length Test Samples and Gap Calibration Samples click on the Set Test Sample button to implement these settings in the spectra table Clear Test Spectra Clicking on this button assigns all spectra to the calibration set Exchange Test lt gt Calibration This function reverses the assignment of the spectra to the calibration and test set Special Setting Spectra with a non specified component value indicated by a blank entry field in the spectra list or with a component value of 0 or 1 must be excluded from the spectra list before you start the validation for the component in question To facilitate the exclusion of those spectra select the component in question and the corresponding option for the component value blank value 0 or value 1 from the corresponding drop down lists See figure 56 Afterwards first click on the Set button and then on the Exit button 66 OPUS QUANT Bruker Optik GmbH Setup Quant 2 Method Spectra Set Spectra on Excluded x All spectra which have no defined component value blank or 1 of O as value for the selected component are set on Excluded Blank T Methanol EE 1 0 Blank Figure 56 Set Spectra on Excluded Automatic Selection of Test Samples This function figure 54 facilitates the assignment of the measured spectra to the calibration set and the test
56. See figure 104 Direct Command Entry E xj Direct Command Entry Send Commands to Optical Bench Send Command Ern MCID Command CONFIG MCID nswer00 00 AD 07 AB Exit Cancel Help Figure 104 Direct Entry Command Dialog Window Bruker Optik GmbH OPUS QUANT 119 Method Protection 2 3 To get the MAC ID via the Internet Explorer enter the Internet address of the spectrometer and navigate to the Service ee View Instrument Configuration page See figure 105 Datei Bearbeiten Ansicht Favoriten Extras degi gt x E a B suchen GaFavoriten Medien E ES E H2 Adresse http 47149 236 30 66 config cfa_etrlerhtm Matrix M SN_MM 0147 04B Instrument Configuration Embedded Web Server EWS15 Firmware Version 1 220 Nov 21 2002 IEWS15 CPU i AMD Elan C400 66MHz Base RAM KB 1632 Extended RAM KB 7168 GO Address in file C EWS TCPIP INI Dec 149 236 30 66 IP Subnet Mask in file C EWS TCPIP INT Dec 255 255 255 0 GATEWAY in file C EWS TCPIP INT Dec 149 236 30 1 Hardware MAC ID Hex 00 00 AD 07 AB ID ITCPIP Settings from CJEWS TCPIP INI Communication Format Code GG EWS DIP Switch 1 OEE EWS DIP Switch 2 OEE EWS DIP Switch 3 OFF EWS15 Board Serial Number 000118 rSCT15 Board Figure 105 MAC ID via Internet Explorer As user A still wants to have access to his method he has to make a copy of the method If t
57. TiQuant Examples CAQUANTiQuant Examples 3076_GMO 3076_GM 1 3077_GM 0 126 _ C QUANTiQuant Examples 3077_GM 1 Figure 38 Quant 2 Multiple File Analysis Add Component Columns Click on the Add Spectra button A standard Load File dialog box opens Select the spectra you want to analyze Upon confirming your selection these spectra are loaded and displayed in a table on the Spec Switch to the Methods page Click on the Add Method button and select one or several methods you want to use for the analysis If you routinely use the same set of methods you can store the set by clicking on the Save 1 tra page 2 Method List button 3 Switch to the Analysis Results page and start the analysis by clicking on the Analyze button The QUANT software will process all files specified on the Spectra page using all methods indicated on the Methods page The results are listed in form of a table on the Analysis Results page You can sort the list according to each column by double clicking on the respective column header To print the analysis results click on the Print button In the field Print Title you can enter a title that will be printed together with the report Bruker Optik GmbH OPUS QUANT 51 Performing a quantitative Analysis Quant 2 Analysis File List x Methods Spectra Analysis Results Graph Statistics Add Methods Load Method List Save Method List Clear
58. U Figure 23 Results for Models Employing 4 and 10 Factors As the quality of the model improves it becomes increasingly difficult to distin guish the errors of prediction judging from these plots only A better way of determining the optimum rank is plotting the RMSECV values versus the rank Switch to the RVMSECV Rank plot Apparently the model improves drastically up to the rank 4 with rank 5 and 6 still giving slightly better predictions How ever ranks higher than 6 barely improve the model and basically represent the addition of fluctuations noise temperature differences of the samples etc which in fact eventually leads to a degradation of the result It also becomes clear that a calculation up to rank 10 would have been sufficient to determine the optimum rank Restricting the calculation to lower ranks saves processing time as the calibration set contains more samples Bruker Optik GmbH OPUS QUANT 33 Validating the Model RMSECY vs Rank Methanol Cross Validation 12 BORO EO EO se EE a EEE ET eee ek esk ek O NWF OOH Ow Oo TOA redu d ZE db r BS A0 AT 12 Ta d 15 16 ZG RMSEC vs Rank Methanol Cross Validation eam eae eG RU CE are Gee FP Gd SEE EE ee EE OR EEE SW OEE EE SEO RERA ONO SS ee TEKE soi mw TSS hv Uh hth Tl Ve e 3 A SB dd B oS JA At 1213 14 15 16 Figure 24 RMSECV Plot and Enlargement of the Plot In the following chapter we will improve the model by restri
59. ZIPI E E Ei R 100 11 9 112 OPUS QUANT Bruker Optik GmbH Rank The rank is number of PLS vectors Regression line The regression line y ax b with a being the slope and b being the offset is calculated using the least squares method Residual The result of a factorization never describes completely the variance of the spectral data matrix and the concentration data matrix The remaining part which is not accounted for by the factorization is called the residual The spectral residual is important for the recognition of outliers The bigger the residual the more likely is the samples an outlier The spectral residual SpecRes is calculated by a summation over all selected frequency points of the difference spectrum SpecRes Ly Gea 11 10 RMSECV Root Mean Square Error of Cross Validation In case of a cross validation the RVSECYV value can be taken as a criterion to judge the quality of the method RMSECV ft u Differ K PRESS 11 11 1 Ms L RMSEE Root Mean Square Error of Estimation The RMSEE value is calcu lated from the SSE sum with M being the number of standards and R the rank 1 RMSEE u RE 11 12 RMSEP Root Mean Square Error of Prediction In case of a test set validation the RMSEP value can be taken as a criterion to judge the quality of the method RMSEP Jatuna 11 13 RPD Residual Prediction Deviation The residual prediction deviation is the ra
60. a tions show a very strong absorption g i Propanol o Methanol 2 50 9000 8500 8000 7500 7000 6500 6000 5500 5000 4500 9000 8500 8000 7500 7000 6500 6000 5500 5000 4500 Ethanol f Mixture i 9000 8000 7000 6000 5000 4000 9000 8000 7000 6000 5000 4000 Figure 8 NIR Spectra of Pure Ethanol Methanol and Propanol as well as a Mixture of these Alcohols Bruker Optik GmbH OPUS QUANT 19 Setting up a Calibration Method The folder Quanttutorial contains 30 spectra acquired from 15 different sam ples Each sample has been measured twice for example O5AIk12 1 and OSAIk12 2 are two spectra taken from the same sample These spectra are to form the calibration set which will be used to perform a cross validation Under real conditions this set would most likely contain much more samples to yield a more robust model 1 Select Setup Quant 2 Method from the Evaluate menu A window with a number of pages opens and the first page Load Method is displayed This page allows you to load an existing Quant 2 method In addition statistical information about the method is displayed To create a new method click on the Components tab Setup Quant 2 Method New x Load Method Components Spectra Parameters Validate Graph Report Store Method Optimize Settings m General information Standards total D Calibration spectra D Test spectra D pr Components 0 Frequency ranges 0 Se
61. a Parameter Graph Store Method Window points 7 b Calculate Transfer Model J Mean Centering Figure 116 Setup Spectra Transfer Method Parameter The graph shows the mean difference between the master spectrum and the cor responding transferred slave spectrum If you want to enlarge a graph detail left click in the graph and draw a frame around the area of interest while pressing the left mouse button To undo the enlargement right click once in the graph Bruker Optik GmbH OPUS QUANT 127 Spectra Transfer Setup Spectra Transfer Method New a Convert 30 JCAMP File to OPUS Files Load Method Spectra Parameter Graph Store Method Display Difference Spectra Display Transferred Spectra Mean Difference between Master and transferred Slave Spectra 0 028 gt 0 002 0 005 115 225 335 445 555 665 7 75 8 85 9 95 10 11 Figure 117 Setup Spectra Transfer Method Graph Moreover you can have the difference spectra and the transferred spectra dis played by clicking on the corresponding button Figure 118 shows the differ ence spectra that have been calculated on the basis of a master spectrum and the corresponding slave spectrum 0 020 0 000 0 020 0 040 9000 8500 8000 7500 7000 6500 6000 5500 5000 4500 Show Filename ST
62. age values minimum maximum 2 must be equal to the wanted sum of component values Enter names for the components the number of samples you need and click on the button Search Component Values Figure 42 Quant 2 Calibration Design Setup Page After clicking on Search Component Values button the QUANT software cal culates an independent sample set more precisely the concentration values for the components These values are listed on the Table page in the dialog box 54 OPUS QUANT Bruker Optik GmbH Calibration Design x Setup Table Graph Print 17 2472 6 6240 66 5651 3 0250 34 9162 57 6666 6 1846 31 1243 60 3867 11 4573 25 3136 55 3868 25 7903 19 2343 46 1513 20 7419 10 3000 64 0068 11 3080 37 9894 49 0860 25 0121 228675 429255 14 1005 42 8861 40 9299 PEIO 39 1690 55 6807 27 1297 2 0844 65 4753 16 1898 32 0307 49 4659 27 9565 5 1149 59 5129 1 2836 47 8484 46 5847 2 7512 35 5571 60 3949 134669 19 5868 62 9902 26 7479 12 0792 55 7137 28 2412 20 8380 48 1594 13 2215 15 0029 62 5571 174691 368007 545039 11 8097 27 3598 56 3234 ei Figure 43 Quant 2 Calibration Design Table The sum of the calculated component concentration values is constant for all samples Normally the sum equals 100 e g the sum of the component concen tration values of a liquid sample is
63. alculation is meaningful only if there is a large number of calibra tion standards because the set should not change significantly when reduced by one or more standards The size of the prediction error is another important number This value can be judged only if the distribution of the component values is known This is taken into consideration in the calculation of R and therefore is a direct measure for the quality of the prediction The relation between R and RMSECYV is not linear as figure 4 shows Ge Diery e DjVu R 100 2 20 10 OPUS QUANT Bruker Optik GmbH 100 90 80 70 R2 60 50 40 30 0 2 0 4 0 6 0 8 1 0 RMSECV Figure 4 R Plotted against RMSECV Bad calibration standards can be recognized by their true values not being predicted using the remaining spectra Using the difference values an automatic outlier detection is performed to mark the samples whose deviation from the true concentration value is particularly large and statistically significant In these cases an FValue is calculated _ M 1 Differ Differ JF FValue 2 21 If the standards are divided up into a set of calibration spectra and a set of test or validation spectra an external validation test set validation can be per formed The calibration is done with the original set of calibration spectra and the test spectra are predicted The mean prediction error is called root mean square e
64. alidate Clicking on the Validate button starts the validation process Prior to the calcu lation you are prompted to enter a name of the validation run Assigning a name to a validation run helps to distinguish between the respective runs while you are optimizing a method The default name is Validation No Confirming this dialog automatically starts the calculation Set alidation Name E x Please enter a name for the validation Validation No1 Cancel Figure 71 Setup Quant 2 Method Set Validation Name Status Bar The status bar informs you about the progress of the calculation Bruker Optik GmbH OPUS QUANT 83 Reference Section 10 6 Setup Quant 2 Method Graph Setup Quant 2 Method C OPUS 6 0 Data Extended Demodata QuantTutorial Alk_d2 q2 m x Load Method Components Spectra Parameters Validate Graph Report Store Method Optimize Settings Validation Validation Nol Prediction True DI Methanol zl Rank E Hsc TG Calibration Re 99 95 0 636 RFD 45 3 Bias 0 00221 RMSECY Iv Line F Color Prediction vs True Methanol Cross Validation _Window Offset 0 122 Slope 0 996 Corr Coeff 0 9998 Figure 72 Setup Quant 2 Method Display of the Predicted Concentrations against the True Values Validation Calibration Choose whether y
65. alysis The OPUS VALIDATION software package is active if the corresponding check box on the 27CFRI1 Rights page in the User Settings dialog box is activated User Settings x Diagnostics Instrument Test Instrument Test General H CRH Rights Preferences Display m User has the right to V Change Parameters V Customize workspace I Edit VBScripts IV Change User Rights and add new Workspaces Validation Options IV Work in validated Environment 21 CFR D M Work in GLP Mode Save original Data Cancel Apply Help Figure 96 User Settings 21 CFR 11 Rights Setup 12 1 Signing Spectra You can only set up a calibration method if the spectra have been signed before with the category Release This measure guarantees that all spectra used for a calibration model can be traced back If you try to validate a method with spec tra which have not been signed beforehand the following error message appears E Setup Quant 2 Method The spectrum file has no proper signature C OPUS 6 0 Data Extended Demodata QuantTutorial OSALK1 Figure 97 Error Message Spectrum File not Signed The procedure of signing spectra files is described in the OPUS VALIDATION Manual chapter 4 4 If all spectra are signed you can perform the validation and further calculations using the OPUS QUANT software Bruker Optik GmbH OPUS QUANT 115 21 CFR part 11 Compliance 12 2 Signing Methods When y
66. an effect on the colored display of the data points on the Graph page If the Color check box in the left lower corner on the Graph page is activated the corresponding data points in the plot are displayed in the color you have specified using this function Copy Spectra The function Copy Spectra copies all spectra of the table and the current method into the directory you have specified Use this function if you want to archive the method together with the spectra used Either enter the path into the corre sponding field or click on Select Path button to specify the target directory In case you have acquired several spectra per sample this function provides an additional opportunity storing a new Quant method which is based on the mean sample spectra See figure 57 To do this OPUS first calculates the mean spectrum for each sample using the full spectral range and the original spectra i e no preprocessed spectra and then creates a Quant method file on the basis of the calculated mean sample spectra This newly created method file is stored in the directory you have defined before by clicking on the Store New Method button and specifying the path in the Store Quant 2 Method dialog window The proposed file name is lt Name of the loaded method gt _AV q2 The mean spectra are saved in the corresponding subdirectory lt Name of new method gt _Spectra The file name including the extension of the mean spectrum is taken over from the first
67. and the file name of the transferred spectra as well as the concentration values of the indi vidual components are entered automatically See figure 126 Note If you have transferred the spectra using the Setup Spectra Transfer Method dialog window see the first procedure described in section 14 2 you need to change the path in the Setup Quant 2 Method dialog window as in this case OPUS has stored the transferred spectra automatically under a different path namely in the self created subfolder Transfer Bruker Optik GmbH OPUS QUANT 133 Spectra Transfer Change Path Set Data Set Copy Spectra Comp Correlations Setup Quant 2 Method New Load Method Components Spectra Parameters Validate Graph Report Store Method Optimize Settings Add Spectra Set Sample Numbers Window Print Dataset Sample Path Fiemame Moisture Protein asn Calibration 1 C TransferiC AHE 8 1 2003 1 13 98 59 05 3 22 13 94 2 Calibration 2 C TransferiC AHE 8 1 2003 2 13 98 59 05 3 22 13 94 3 Calibration 3 CATransteriC AHE 16 1 2003 1 9 74 65 41 3 58 10 7 4 Calibration 4 Geneen AHE 16 1 2003 2 9 74 65 41 3 58 10 7 Calibration 5 C MTransferiC AHE 17 1 2003 1 12 74 5416 2 78 19 88 e Calibration 6 C MTransteriC AHE 17 1 2003 2 12 71 5416 2 78 19 88 Calibration
68. ant principal components will then be used instead of the original spectral data thus leading to a considerable reduction of the amount of data A PLS regression algorithm will be deployed to find the best correlation function between spectral and concentration data matrix The determination of the number of principal components is a crucial point for the quality of the calibration model Using an insufficient number of principal components leads to a poor reproduction of the spectral data and therefore the model will not be able to recognize changes in the spectral features This is called underfitting On the other hand including too many principal compo nents just adds spectral noise to the regression and does not increase the amount of valuable information overfitting Multicomponent systems can be analyzed either for each component separately PLS 1 algorithm or simultaneously for all components PLS 2 algorithm However the PLS 1 analysis usually yields better results and therefore is mainly used for multivariate calibrations QUANT exclusively uses the PLS 1 algorithm Details about the theory behind the multivariate calibration and its implementation in QUANT are described in chapter 2 4 OPUS QUANT Bruker Optik GmbH Theoretical Background In general the aim of a quantitative analytical method is to determine the prop erty Y of a system from an experimentally observable X whereby X and Y are correlated by
69. art of the infor mation of the data set is represented by only one factor As many factors are cal culated as there are spectra in the data set The factors are calculated for the whole data set In case you change the data set or the parameters e g frequency region data preprocessing the factors have to be calculated once again The scores contain the information about how the original spectra are described by factors For each spectrum there is a set of scores that describes the spectrum on the basis of the calculated factors By multiplying the score coefficients by the corresponding factors and adding up the products a spectrum can be recon structed The scores can be used for further evaluations as they represent the spectral information of the original spectra on the basis of the loadings Use only the score values of the first factors as the higher factors represent only noise and other non usable information The principle of the factor analysis is illustrated on basis of the following exam ple A data set consisting of five simple spectra that have two overlapping bands are factorized See figure 67 The calculated factor 1 includes the biggest part of the information principle variance of the data set The result of factor 1 is a factor spectrum that is similar to an mean spectrum Factor 2 includes the information about the two varying bands To determine which band is more 78 OPUS QUANT Bruker Optik GmbH Setup Quant 2
70. as yielded a RMSECV value of 0 636 for the methanol concen tration which is reasonable considering the fact that the whole frequency region of the spectra has been used including the spectral noise as well as the region showing total absorption The next chapter shows you how the model can be improved But let us have a closer look at the validation results first Switch back to the first graph Prediction True The straight line represents a prediction without any error that is the predicted concentration values equal the concentration val ues of the test samples Now enlarge a part of the graph by left clicking in the graph and drawing a frame around the area of interest As you can see the pre dicted values lie close to the line but not all of them actually match the line Prediction vs True Methanol Cross Validation 110 100 90 80 70 60 50 40 30 0 5 10 20 30 40 50 60 70 SU BU 100 Prediction vs True Methanol Cross Validation 18 21 22 23 24 25 26 27 26 29 30 31 32 33 34 35 36 3 38 Figure 21 Enlarging a Region of the Graph Bruker Optik GmbH OPUS QUANT 31 Validating the Model Select the respective components from the drop down list to display the results for the other components Note that the recommended rank remains the same although the RMESCV values are different Now display the result for the first rank by se
71. ata Block AB Compounds total 3 Frequency Regions 1 Selected Datapoints 961 Droanracacecine Qanand Darivative Figure 33 Setup Quant 2 Method Validation Report View 4 To print this report you can click either on the Print or on the Window button Clicking on the Window button embeds the QUANT setup assis tant in an OPUS window figure 34 The same Window button for the QUANT Setup Assistant is also on the Graph and Spectra page Then choose the Print command from the OPUS Print menu If you want to copy the whole report to the clipboard mark the report by clicking on the upper left tile in the validation report see figure 33 and press Ctrl C on the keyboard Then you can paste the content of the clipboard into any other application e g Microsoft Word Notice that upon embedding the QUANT Setup Assistant into OPUS a control panel consisting of three buttons becomes active figure 34 Clicking on one of these buttons brings you back to the respective page of the Setup Quant 2 Method dialog window Click on the Report button to return to the Report page Bruker Optik GmbH OPUS QUANT 45 Generating a Report and Saving the Method b Quant Report full_access ows 2 Operator Default Administrator i File Edit Heu Window Measure Manipulate Evaluate Display Print Macro Validation Setup Help hE dk ZO d E Gu SE B e PS IE E E E Print Spectra Ht E ZG Sj d E OPUS Browser ax T
72. ation 1 1 1 0 9 0 6 D 0 01 0 02 0 03 0 04 0 05 0 06 0 07 0 08 Figure 3 Mahalanobis Distance Plotted against the Spectral Residual Do not be deceived by good results from a calibration particularly at high ranks Since the spectra and the component values are present as input it is not difficult to reproduce the component values Fit True using enough PLS vec tors This fact is completely different than the prediction of a sample which is not contained in the calibration set as it is done in the validations In case of a cross validation the root mean square error of cross validation RMSECY can be taken as a criterion to judge the quality of the method _ fl geo 5 RMSECV ae wer ap PRESS 2 18 Bruker Optik GmbH OPUS QUANT 9 Theoretical Background In case of a test set validation this value is called the root mean square error of prediction RMSEP M PRESS X Differ 2 19 A recommendation for the optimal PLS rank is given using these values to pre vent overfitting 1 The rank with the smallest PRESS value is searched This presumes that enough PLS ranks are calculated 2 For all lower ranks the quotient of their PRESS values and the minimum is calculated FValue 3 From this FValue a probability is calculated FProb FValue M M 4 The rank having a probability smaller than 0 75 for the first time is marked as the optimum rank The PRESS c
73. board into the table See figure 125 132 OPUS QUANT Bruker Optik GmbH Setting up a Quant 2 Method using transferred Spectra Quant Report full_access ows 2 Operator Default Administrator _ e File Edit View Window Measure Manipulate Evaluate Display Print Macro Validation Setup Help GE GZ AEC LE T s eee ZERE GZ wp ax erator Default Adminis 2 Operator Default Aj Go back to i Graph Report For Help press F1 CATranster C AHE 8 1 2003 1 13 98 Figure 125 Pasteing Spectra plus Concentration Values C MransteriC AHE 8 1 2003 2 13 98 59 05 322 13 34 1 03 C MransteriC AHE 16 1 2003 1 3 74 Jesa 358 DO PZO C MransteriC AHE 16 1 2003 2 9 74 Jesa 358 DO 0 95 C MransteriC AHE 17 1 2003 1 12 71 5416 278 19 68 115 CATransfert AHE 17 1 2003 2 12 71 54 16 278 19 88 115 C Mransteric AHE 18 1 2003 1 10 76 64 56 3 28 10 65 122 C MransteriC AHE 18 1 2003 2 10 76 64 56 3 28 10 85 1 22 CATransteri AHE 20 1 2003 1 16 45 56 38 2 79 14 03 0 99 C ATransferic AHE 20 1 2003 2 16 45 56 38 279 1409 0 99 Gere AHE 21 1 2003 1 13 63 5879 3 03 14 04 1 C ATransferic AHE 21 1 2003 2 13 63 5879 3 03 14 04 D C AMTransferic AHE 22 1 2003 1 13 59 59 57 3 04 13 48 jos Gere AHE 22
74. button The progress of the opti mization will be displayed in the status bar Use the OPUS task bar to stop a running optimization process See figure 87 98 OPUS QUANT Bruker Optik GmbH Setup Quant 2 Method Optimize x Use Parameters Methanol DI NIR ba Number RMSECY Rank Regions Preprocessing 1 0 1 7 12001 7 7497 2 No Spectral Data Preprocessing 2 0 196 10 7501 1 6097 3 No Spectral Data Preprocessing 3 0 101 8 12001 7 6097 3 No Spectral Data Preprocessing 4 0 153 6 6101 1 5449 4 No Spectral Data Preprocessing 5 0 133 6 12001 7 7497 2 6101 1 5449 4 No Spectral Data Preprocessing 6 0 139 7 7501 1 5449 4 No Spectral Data Preprocessing 7 0 148 7 12001 7 5449 4 No Spectral Data Preprocessing 8 0 908 4 5453 2 4597 1 No Spectral Data Preprocessing 9 0 808 8 12001 7 7497 2 5453 2 4597 1 No Spectral Data Preprocessing 10 0 236 g 7501 1 6097 3 5453 2 4597 1 No Spectral Data Preprocessing 11 0 181 9 12001 7 6097 3 5453 2 4597 1 No Spectral Data Preprocessing 12 0 177 8 6101 1 4597 1 No Spectral Data Preprocessing 13 0 151 8 12001 7 7497 2 6101 1 4597 1 No Spectral Data Preprocessing 14 0 191 7 7501 1 4597 1 No Spectral Data Preprocessing 15 0 222 6 12001 7 4597 1 No Spectral Data Preprocessing 16 10 3 1 4600 9 4250 No Spectral Data Preprocessing 17 2 93 7 12001 7 7497 2 4600 9 4250 No Spectral Data Preprocessing 18 3 46 6 7501 1 6097 3 4600 9 4250 No Spectral Data Prep
75. cient number of samples to the calibration set covering the new wider range In case you prepare the samples for your calibration set in the lab make sure that these samples show no collinearity which means that they do not show a linear de or increase in concentration of the com ponents Especially dilution series are not suited as calibration sam ples Bruker Optik GmbH OPUS QUANT 15 Chemometric Models and their Validation 1 A 2 A 3 A 2 B 4 B 6 B 3 C 6 C 9 C 1 A 5 A 13 A 2 B 12 B 7 B 3 C 1 C 24 C Figure 7 Example of collinear Samples and Samples showing no Collinearity e When acquiring spectra from the calibration set never measure the samples in increasing or decreasing order of their concentration Oth erwise linear fluctuations in temperature heating up or cooling of the samples or concentration evaporation of solvent will not be recognized by the PLS 1 algorithm If possible repeat the measure ments at a later point in time e Ensure that the reference method you use for the determination of the components concentration yields reliable results Repeat these mea surements to obtain statistical significance Be sure to know the sta tistical error of your reference method 3 2 Acquiring Spectra and Data Preprocessing After you have chosen a set of samples you need to acquire their IR spectra Check the reproducibility of the measurements for short and long time inter va
76. cting the frequency region and performing a data preprocessing 34 OPUS QUANT Bruker Optik GmbH Improving the Model The first step in improving a chemometric model is to focus the PLS regression on groups that contain information significant for the system From the explana tion in chapter 4 you have learned that the region below 4400cm does not con tain any useful spectral information as noise prevails The peak around 4800 cm shows a very strong absorption and should therefore also not be included In addition you should limit the frequency region to 9000cm because above this value there are not any spectral information 1 To repeat the validation you need not set up the entire Quant method once again Just switch to the Parameters page and change the fre quency range limits 2 You can specify the frequency range limits by either entering the values into the table or clicking on nteractive Region Selection button Setup Quant 2 Method C OPUS 6 0 Data Extended Demodata QuantTutorial Alk dz A xj Load Method Components Spectra Parameters Validate Graph Report Store Method Optimize Settings e Preprocessing in individual regions PS Set m Preprocessing in calibration regions No spectral data preprocessing DI V Mean Centering rm Calibration regions Fa a E 12001 7 Interactive Region Selection Clear Selected Regions m View s
77. ctrometer and the computer has been established If you have modified a protected method you can store the method file either under the same or a different file name However the protection status cannot be changed If you have worked with a protected Quant 2 method and then want to use an unprotected method first exit the Setup Quant 2 Method dialog window open it again and load the unprotected method file 124 OPUS QUANT Bruker Optik GmbH Setting up a Spectra Transfer Method 1 4 Spectra Transfer The OPUS software allows you to adapt foreign spectra i e spectra you have acquired using a spectrometer system from another manufacturer or a Bruker spectrometer but with a different accessory to those OPUS spectra with which you want to work e g setting up a Quant 2 method Before starting the spectra transfer you first have to set up a spectra transfer method The purpose of this method is to model the differences between the OPUS spectra and the original spectra caused by the different spectrometer systems During the subsequent spectra transfer these differences are taken into consideration and the original spectra are adjusted correspondingly Note The calculation of the spectra transfer model is based on the PDS method Piecewise Direct Standardization Before setting up a spectra transfer method you first have to measure a sample set consisting of approximately 15 to 20 samples using both spectrometer sys te
78. d The algorithm of this Function is licensed by University of Washington Second Order Instrument Standardization US Patent Number 5 459 677 developed by Dr s Bruce R Kowalski and Yongdong Wang Transfer Cancel Help Figure 121 Spectra Transfer 130 OPUS QUANT Bruker Optik GmbH Setting up a Quant 2 Method using transferred Spectra 14 3 Setting up a Quant 2 Method using transferred Spectra After you have transferred the original spectra successfully to OPUS you can use them for setting up a Quant 2 method To add the transferred spectra and the corresponding component values of the individual components to the spectra list figure 126 in a time saving manner proceed as follows e Select in the Evaluate menu the Setup Quant 2 Method function and click in the Setup Quant 2 Method window on the Components tab Enter the names of your components See figure 122 Setup Quant 2 Method New k xi Load Method Components Spectra Parameters Validate Graph Report Store Method Optimize Settings Add Component Name Unit Crude Fat mg Formatting in the Quant 2 analysis report Monas Imo l Default settings 5 significant digits ASH mo l mg Digits after the decimal point Crude Fiber Figure 122 Entering Component Names e In the directory of the 3D JCAMP file a text file lt File Name gt _info txt has been created This file contains the path and
79. dition to the method The result comprises the graphs and the reports Store Method If you click on the Store Method button the Select Validation Results dialog box figure 84 opens Select the validation s from which you want to store the results In addition the last Optimization run will be saved as well This allows you to perform time consuming optimizations during off hours and save the results afterwards Click on the Select All button to automatically select all vali dations listed Select Validation Results x The results of the selected validations will be stored Select All Validation No 2 Validation No 3 Validation No 4 Validation No 5 Cancel Figure 84 Setup Quant 2 Method Saving the Method Bruker Optik GmbH OPUS QUANT 95 Reference Section Component Table This table lists all the components used by the selected method For each com ponent the recommended rank R RMSECV and the rank used for the QUANT analysis are listed The column Rank Method lists the recommended rank as default but this value can be edited by the user These values will be used for the QUANT analysis later on Using the check boxes in the last column you can specify whether a component will be used for the analysis or not Spectral Residuals If you activate this check box the spectral residuals will be calculated during the analysis Factor for Mahalanobis Distance Limit The factor for the Mahalanob
80. e RMESCV value 0 139 has further improved while the optimum rank is still 7 38 OPUS QUANT Bruker Optik GmbH Setup Quant 2 Method C OPUS 6 0 Data Extended Demodata QuantTutorial Alk_d2 q2 Load Method Components Spectra Parameters Validate Graph Report Store Method Optimize Settings Validation Validation No3 DI dE Calibration Prediction True x Methanol Rank E DI Rec 7 Special Re Window 100 Print RMSECY 0 139 Save RPD 209 Bias 0 0191 Spectra Loadings I Line Figure 29 Result of the Validation when Using the Second Derivative of the Spectra No general recommendations can be given as to which data preprocess ing method should be used The best method has to be found empirically by trial and error The following table compares the validation results for different preprocessing methods applied to different frequency regions Table 1 Comparison of Different Data Preprocessing Methods Validation Gerea Gezi Rank RMSECV 1 none 9000 5300 7 0 167 2 Straight Line 9000 5300 6 0 155 3 First Derivative 9000 5300 6 0 147 4 Second Derivative 9000 5300 7 0 120 5 none 7600 5300 7 0 164 6 First Derivative 7600 5300 5 0 163 7 Second Derivative 7600 5300 6 0 147 Bruker Optik GmbH OPUS QUANT 39
81. e calibra tion Moreover the plot displays the standard deviation for each sample indi cated by a blue cross The red line indicates the mean predicted component value of each sample Rank The value of the rank displayed in the graph The default value is the recom mended rank Rec The recommended rank found during the validation of the method R2 The value of the coefficient of determination B for the rank displayed in the graph You can conveniently browse the R values for different ranks by plac ing the cursor in the rank field and using the arrow keys of your keyboard to change the rank RMSECV RMSEP The value of the root mean square error of cross validation RMSECYV or the root mean square error of prediction RMSEP respectively for the rank dis played in the graph RPD The value of the residual prediction deviation for the rank displayed in the graph Bias The bias value for the rank displayed in the graph Line In case of the Prediction True plot the Line check box is available If you acti vate this check box the regression line blue line is drawn into the plot The off set and the slope of the regression line depend on the selected rank Below the plot the exact values of the offset and the slope of the regression line as well as the correlation coefficient value are displayed These values depend on the selected rank Note In chemometrics a large number of statistical parameters is used
82. e reve Calibration 1 COPUS 6 0Dat O54LK11 0 100 Calibration 1 COPUS 6 0Dat 05ALK1 2 gt 0 100 3 Calibration 2 CAOPUS 6 0 Dat 05ALK21 100 0 0 Calibration 2 CAOPUS 6 0 Dat O54LK2 2 100 0 0 5 Test 3 CAOPUS 6 0 Dat O5ALK3 1 o 100 o 6 _ Test 3 C OPUS 6 0 Dat O5ALK3 2 Q 100 0 Test A CAOPUS 6 0 Dat 05ALK41 133 364 33 278 33 356 e Test 4 CAOPUS 6 0Dat 05ALK4 2 33 364 33 278 33 356 3 Calibration 5 CAOPUS 6 0 Dat 05ALK5 1 49 666 25 435 24 899 140 Calibration 5 COPUS 6 0 Dat O54LK5 2 49 666 25 435 24 899 41 Calibration 6 C OPUS 6 0 Dat O54LK6 1 24 942 24 982 50 078 42 Calibration 6 CAOPUS 6 0 Dat O54LK6 2 24 942 24 982 50 078 43 Test 7 C AOPUS 6 0 Dat O5ALK7 1 26 392 48 95 24 658 Test 7 CANOPUS 6 0 Dat 05ALK7 2 26 392 48 95 24 658 a5 Test 8 C AOPUS 6 0 Dat O5ALK8 1 50 017 i 49 983 16 _ Test 8 C AOPUS 6 0 Dat 054LK8 2 50 017 0 49 983 Calibration g CAOPUS 6 0Dat 05ALK9 1 66 648 33 352 0 18 Calibration E CAOPUS 6 0 Dat O54LK9 2 66 648 33 352 0 149 Calibration 10 CAOPUS 6 0 Dat O54LK10 1 0 33 392 66 606 20 Calibration 10 GORE 6 0 Dat 05ALK10 2 0 33 392 66 606 Figure 55 Alternating Calibration Set and Test Set Spectra Leave Excluded Spectra If you activate the Leave Exclude Spectra check box the spectra specified as Excluded in the spectrum table will not be assigned to another data set i e they remain excluded Set Test
83. e root of this sum This method is used to account for different samples thickness for exam ple e Min max Normalization first subtracts a linear offset and then sets the y maximum to a value of 2 by multiplication with a constant Used similar to the vector normalization e Multiplicative Scatter Correction performs a linear transformation of each spectrum for it to best match the mean spectrum of the whole set This method is often used for spectra measured in diffuse reflec tion e First Derivative calculates the first derivative of the spectrum This method emphasizes steep edges of a peak It is used to emphasize pronounced but small features over a broad background Spectral noise is also enhanced e Second Derivative similar to the first derivative but with a more drastic result Bruker Optik GmbH OPUS QUANT 17 Chemometric Models and their Validation No general recommendation can be given whether a given data set should be preprocessed or which method is suited best for it Therefore the optimal data preprocessing method can only be found empirically by applying several meth ods to your spectral data and comparing the results 3 3 Validating the Model At this point the model needs to be validated If a sufficient number of samples have been measured it is possible to divide the samples into two sets of about equal number a calibration set and a test set The calibration set is used to build up a model which
84. eady mentioned it is the responsibility of the user to choose the optimal parameter set Repeat the validation using the selected parame ter set To do this click on the corresponding row and then on the Use Parameters button As a result these parameters are automatically pasted into the respective fields on the Parameter page and the software switches to the Validate page On the Settings page you can restrict the maximum frequency region and select the data preprocessing methods used for the optimization 40 OPUS QUANT Bruker Optik GmbH GO GO rd ZO GI d Gk D rk 0 7 1 8 6 6 7 E A 8 9 9 8 8 7 6 1 E 6 a 12001 7 7497 2 7501 1 6097 3 12001 7 6097 3 6101 1 5449 4 12001 7 7497 2 6101 1 5449 4 7501 1 5449 4 12001 7 5449 4 5453 2 4597 1 12001 7 7497 2 5453 2 4597 1 7501 1 6097 3 5453 2 4597 1 12001 7 6097 3 5453 2 4597 1 6101 1 4597 1 12001 7 7497 2 6101 1 4597 1 7501 1 4597 1 12001 7 4597 1 4600 9 4250 12001 7 7497 2 4600 9 4250 7501 1 6097 3 4600 9 4250 12001 7 RNG ASTO APRN Setup Quant 2 Method C OPUS 6 0 Data Extended Demodata QuantTutorial Alk d2 q2 No Spectral Data Preprocessing No Spectral Data Preprocessing No Spectral Data Preprocessing No Spectral Data Preprocessing No Spectral Data Preprocessing No Spectral Data Preprocessing No Spectral Data Preprocessing No Spectral Data Preprocessing No Spectral Data Preproces
85. ent values are OK The variable Factor was introduced for setting a more realistic limit for the outlier detection during analysis 0 40 0 35 0 30 0 25 Lever 0 20 0 15 0 10 0 05 4 6 8 10 12 14 16 True Figure 2 Leverage Values Plotted against the True Values The measured calibration spectrum after the data preprocessing is represented by x and the spectrum reconstructed from the PLS vectors v as s t are the score coefficients SS Lili 2 14 The spectral residual SpecRes is calculated by a summation of all selected frequency points of the difference spectrum SpecRes KN DEAK 2 15 8 OPUS QUANT Bruker Optik GmbH The better the reproduction of a spectrum is the smaller is the spectral residual To recognize outliers the squared spectral residual is compared with the mean value of all others by calculating the FValue using the following formula LM 1 SpecRes gt i SpecRes Tzi FValue 2 16 Spectra poorly represented by the PLS vectors have a high FValue From the FValue and the number of degrees of freedom a probability FProb can be calcu lated FProb indicates the probability that a standard is a spectral outlier The limit for the automatic outlier detection is 99 If the FProb value lies above the limit the corresponding spectrum is indicated in the report by a grayed line FProb FValue 1 M 1 gt 0 99 2 17 Mah Dist vs Spec Residual Test Set Valid
86. er ence of both values Setup Quant 2 Method C OPUS 6 0 Data Extended Demodata QuantTutorial Alk x Load Method Components Spectra Parameters Validate Graph Report Store Method Optimize Settings Validation C Calibration Validation No3 Mi True Prediction hd Rank b 7 Rec 7 _ Filename True Prediction Difference Window 4 osakin O 0 2265 0 226 05ALK1 2 D 0185 0185 05ALK21 100 99 83 0 166 Print a 05aLK2 2 100 99 69 0 306 ZIEGA 0 oos633 0 0663 6 osaks2 0 0 09495 0 0949 O5ALK4 1 33 364 33 12 0 245 EEE SISA 33 364 33 42 0 0584 9 fosaLk51 49 666 49 65 0 0169 40 054LK5 2 49666 49 67 0 00256 O5SALK6 1 24 942 24 97 0 0313 OSALKB 2 24 942 ETZE 0 0221 O5SALK7 1 26 392 26 46 0 0683 M4 O5ALK7 2 26 392 26 41 0 0216 45 O5aLKa1 50 017 50 01 0 00948 46 O54LK8 2 50 017 50 0 0164 05ALK9 1 66 648 66 79 044 05ALK9 2 66 648 66 66 0 0159 zj Figure 31 Setup Quant 2 Method Report Page Bruker Optik GmbH OPUS QUANT 43 Generating a Report and Saving the Method 3 Now select Validation Report from the drop down list figure 32 The view changes instead of the table a print ready validation report is dis played See figure 33 Setup Quant 2 Method C OPUS 6 0 Data Extended Demodata QuantTutorial Alk_d2 q2 E Validation No 3 E 2 True Prediction E Methanol
87. est set 14 18 27 63 65 67 Test set spectrum 65 Test set validation 10 11 13 14 18 27 40 63 65 82 85 98 114 Test spectrum 11 76 Transferred spectra 130 134 True Prediction 43 93 U Underfitting 4 18 Univariate calibration 3 User ID 117 Vv Validation 13 18 27 28 74 82 83 84 92 96 Validation report 44 45 94 Validation result 47 60 91 96 Vector normalization 17 WwW Window 70 87 108
88. esults f Graph Statistics use landscape Print Protein Protein is T ZE Protein 26 0 297 l 0431 Gaz 0 125 15 54 0 313 Jr Protein 26 0 309 0 115 5 2 0 765 0 943 bei Protein 26 0445 UZ 13 36 0 504 ur Oil 26 0 537 0 00685 5 44 1 538 Lr Join 26 Su Dz 435 MA 0 988 oil 26 m 59 0 0767 4 99 9 990 US oil 26 0 959 Ie C QUANT Sunflower Seeds Ga EO EE G QUANT Sunflower Seeds Oil 4 d E Figure 95 Simultaneous Evaluation of several Quant 2 Methods Statistics Page Print Click on this button to print out the statistics in landscape format Bruker Optik GmbH OPUS QUANT 109 Reference Section 110 OPUS QUANT Bruker Optik GmbH 1 1 Abbreviations and Formulas Bias mean value of deviation also called systematic error The bias is a systematic deviation of the measured predicted values from the true value due to a particular measurement method for example In our case it is the differ ence between the average true value and the average measured value of the val idation set samples gt Differ Bias 11 1 Se 11 1 Calibration function The calibration function b correlates a property Y of a system with an experimentally observable X Y X 5 11 2 The vector Y consists of the component values of one component of the refer ence measurements The row vectors of the matrix X are formed from the cali bration spectra T
89. fea True Prediction RAMSECY Concentration Outlier Validation Report a l l b D 5ALK1 1 oF GOIZ fo SALK21 100 GS 0 186 osaLk22 fioo SE oF oS GO GO E SALK3 1 0 0663 0 09495 rS SALK4 1 3 364 3312 OSALK42 33 364 3342 0 5ALK5 1 _ 49 666 gg 0 0169 ALKS 2 49666 OS 0 00256 5ALK6 1 0 0313 OSALK6 2 24 942 SS OSALK7 1 26 392 Zaz 0 0683 GAUR 26 392 SA sake 50 017 Sd Pg OSALKB 2 50017 so EIS ZEAK 66 648 SS 0 14 OSALKS 2 66 648 SEE 0 0159 bam KA ee er GO GO E ojo on Ge Se Se Ge See Ga Zo Ge e cs E GIE b KZ Gira KZ Figure 32 Setup Quant 2 Method Switching to the Validation Report View For each component of your multicomponent mixture a separate report can be generated These reports contain general and specific information about the selected component as well as the used frequency region 44 OPUS QUANT Bruker Optik GmbH Setup Quant 2 Method C OPUS 6 0 Data Extended Demodata QuantTutorial Alk_d2 q x Load Method Components Spectra Parameters Validate Graph Report Store Method Optimize Settings Validation Validation No3 DI GEZA Calibration Methanol DI Rank E DI Rec 7 Validation Report Window Validation Report Pin General Information Exclude Outliers Method File Alk_d2 q2 Standards total 30 Calibration Spectra 30 Test Spectra D D
90. g a certain preprocessing method can be applied to a different or a larger or a smaller frequency region than for the preprocessing in the calibration regions To specify a preprocessing sequence select the desired preprocessing method from the drop down list and define the frequency region by either entering the values in the corresponding field or selecting the region interactively See description below section Interactive Region Selection Then click on the Add to List button You can repeat this procedure several times If you want to mod ify an item in the list later mark this item by clicking on the corresponding number select a different method and or specify another region and click on the Modify Selected Item button To delete the complete list click on the Clear List button Note Do not use a Quant 2 method created with this option in a previous OPUS version Version 5 5 or lower Preprocessing in Calibration Regions This drop down list contains several data preprocessing methods For detailed information about the data processing methods refer to chapter 1 The selected data preprocessing method is applied to the specified calibration region s Mean Centering If you activate this check box the mean spectrum and the mean component val ues are subtracted before the PLS model is performed This scaling is advanta geous Choose it in almost all cases Bruker Optik GmbH OPUS QUANT 73 Reference Section Ca
91. g switch to the Vali date page and start the validation 6 The second validation run yields a rank similar to the one of the first val idation run but the RMSECV value has improved to 0 197 Further more looking at the RMSECV Rank plot a prominent minimum can be observed Setup Quant 2 Method C OPUS 6 0 Data Extended Demodata QuantTutorial Alk_d2 q2 x Load Method Components Spectra Parameters Validate Graph Report Store Method Optimize Settings Validation Validation No2 DI eg Calibration Prediction True z Methanol Rank 7 Rec 7 R j Window 100 Print RMSECY 0 197 Save RPD 147 Bias 0 018 Spectra Loadings E de Special I Color Figure 27 Validation Result after Limiting the Frequency Region Bruker Optik GmbH OPUS QUANT 37 Improving the Model ae 1 8 1 6 1 4 1 2 0 8 0 6 0 4 0 2 RMSEC vs Rank Methanol Cross Validation 0 11562253535 445555 665 7 75 665 9 95 10 Figure 28 Result of Validation after Limiting the Frequency Range RMSECV vs Rank 7 In the next step apply a preprocessing routine to the data prior to the validation run To do this switch to the Parameter page and select Sec ond Derivative 17 smoothing points from the drop down list Do not change the frequency region Now start another validation run As you can see th
92. g the foreign spectrometer are the slave spectra Load these spectra by clicking on the corresponding button See figure 115 Note The master spectra have to be available in the data block Absorbance or Log Reflectance whereas the slave spectra can also be available in the data blocks Transmittance or Reflectance To set up an usable spectra transfer method it is of crucial importance that the spectra of the individual samples are sorted in both tables master and slave in the same order Otherwise you have to rearrange the spectra by selecting the spectrum spectra in question and moving it them to the new position while pressing the left mouse button Note In both tables the spectra are loaded in alphabetic order according to their file names Take this fact into consideration when specifying the file names for the master spectra and slave spectra during the measurement Click on the Parameter tab Normally you can take the default setting window point 7 In case of a shift between the master spectra and the slave spectra with regard to the x axis frequency shift however enter a higher window point value To find out whether the model yields better results with or without Mean Centering give it a try Now click on the Calculate Transfer Model button As a result of this the window switches automatically to the Graph page Setup Spectra Transfer Method New B Convert 3D JCAMP File to OPUS Files Load Method Spectr
93. he frequency region tested during an optimization This is done on the Settings page Also on this page you can specify that the optimization is to run in the background Activate the Run Optimization in Background check box if you want to continue working with OPUS during the optimization Otherwise the OPUS software is blocked during the optimization process Component After you have performed an optimization once you can select the result for a component from the component drop down list Optimization Type The options are MIR suited for NIR data only and General A or B suited for both MIR and NIR data Select the appropriate optimization type from the drop down list NIR If you select this optimization type a set of five frequency regions is used The frequency regions are typical for NIR applications The five frequency regions are tested on their own and in all possible combinations General A The selected frequency region selected on the Settings page is divided into 10 equal subregions To find the optimum combination the calculation starts with 10 subregions and successively excludes one subregion This procedure contin ues until the mean prediction error value does not improve further Bruker Optik GmbH OPUS QUANT 97 Reference Section General B The selected frequency region selected on the Settings page is divided into 10 equal subregions To find the optimum combination the calculation starts with o
94. he original method is stored in the Quant _ Methods directory for example create a subdirectory for user B If several users i e spectrometers are to get a protected method create several subdirectories See the following figure Sd Quant_Methods CJ USER_B_00_00_AD_07_AB_11 USER_C_00_00_4D_ 01 18 11 Figure 106 Creating Subdirectories Copy the method file q2 or faa from the original directory to the new one using the Explorer program of the operating system You can also use the Store Method button in the Setup Quant 2 Method dialog box e Store Method page User A must have an OPUS registration which gives him access to the OPUS package VALIADTION as a method can only be protected in conjunction with Add Signature command To view the available pack ages select in the OPUS Help menu the About OPUS item The follow ing window opens 120 OPUS QUANT Bruker Optik GmbH 4 fond OPUS Version 6 0 Build 6 0 62 413 Copyright Bruker Optik GmbH 1997 2005 This Version of OPUS was licensed to Software Bruker Optik GmbH Workstation 1354984249 3524656654 Key confirmed Available Packages 3D a ADIO ATAB ATR EAI Parts of this Software are based in part on the work of the Independent JPEG Group Figure 107 About OPUS Dialog Box To protect a method select in the OPUS Validation menu the Methods Add Signature Show History function The dialog box shown in figure 1
95. he solution of the above system of equations is given by A xy Oe 11 3 Correlation Coefficient The correlation coefficient is a measure of the linear relation between variables i e the correlation coefficient value indicates how much of a change in one variable is explained by a change in the other variable The correlation coefficient ranges from 1 0 to 1 0 A correlation coefficient of 1 0 or 1 0 indicates a perfect positive or negative relationship in which high values of one variable are related perfectly to high values of the other vari able and conversely low values of one variable are perfectly related to low val ues of the other variable A correlation coefficient of 0 means that there is no linear relation between the variables Differ The difference between the true concentration of a sample i as deter mined by another method and the predicted concentration Differ y SEE 11 4 Factor The concentration data matrix and the spectral data matrix are broken down into pairs of scores and loadings vectors by the PLS algorithm Each of these pairs are called a factor Bruker Optik GmbH OPUS QUANT 111 Abbreviations and Formulas FValue and FProb To recognize outliers the squared spectral residual is com pared with the mean value of all others by calculating the FValue using the fol lowing formula 2 M 1 SpecRes FValue JU GEA a SpecRes j i 11 5 Spectra poorly represented by
96. inates the process Display full_access ows Operator Default Administrator EO File Edit view Window Measure Manipulate Evaluate Display Print Macro Validation Setup Help GE BA ZE dk GES RZ SH See Bes eae d HS Se OPUS Browser For Help press F1 ax 4 Setup Quant 2 Method C OPUS 6 0 Data Extended D EIZIE ax Load Method Components Spectra Parameters Validate Graph Report Store Method Optimize Settings Use Parameters 0 43 23 0 409 24 1 04 25 0 887 26 1 29 27 0 857 28 0 966 29 1 04 30 0 563 31 0 504 32 0 101 33 0 171 34 0 0801 35 0 113 36 0 176 37 0 163 38 0 122 39 0 879 8 HANNA ANAKA Methanol ZG Number RMSECY Rank Regions 22 7501 1 5449 4 4600 9 4250 12001 7 5449 4 4600 9 4250 5453 2 4250 12001 7 7497 2 5453 2 4250 7501 1 6097 3 5453 2 4250 12001 7 6097 3 5453 2 4250 6101 1 4250 12001 7 7497 2 6101 1 4250 7501 1 4250 12001 7 4250 12001 7 7497 2 7501 1 6097 3 12001 7 6097 3 6101 1 5449 4 12001 7 7497 2 6101 1 5449 4 7501 1 5449 4 12001 7 5449 4 5453 2 4597 1 NIR Optimize Preprocessing No spectral data preprocessin No spectral data preprocessin No spectral data preprocessinn No spectral data preprocessirn No spectral data preprocessirn No spectral data preprocessirn No spectral data preprocessin No spectral data preprocessin No spectral data preprocessin
97. ined in the calibration samples has to be known e g it has to be determined by a different analytical technique Then the height of a peak char acteristic for the substance is determined from the spectra and plotted versus the known concentrations The resulting graph will be used to evaluate the concen tration of an unknown sample by measuring the peak height and reading the corresponding concentration from the graph In order to analyze multicompo nent samples a signal characteristic for each component must be used for the calibration and analysis These signals must be well separated to be indicative Univariate calibrations suffer from the following disadvantages e Outliers or perturbations caused by additional unknown components are not recognized because the concentration of the analyte is deter mined in one spectral point only e Statistical fluctuations caused by detector noise are directly reflected by the concentration values Therefore measurements have to be repeated several times e Peaks used for the analysis of multicomponent systems must be well separated which is a severe drawback in NIR spectroscopy e The analysis of multicomponent systems assumes the validity of the Lambert Beers law i e a linear correlation between the concentra tion and the spectral response This does not account for temperature fluctuations or intermolecular interactions Bruker Optik GmbH OPUS QUANT 3 Introduction to multivariate Ca
98. ion vs True Methanol Cross Validation 48 8 49 25 49 45 49 65 49 86 60 05 60 25 Figure 76 Zoomed area enlarged To zoom exactly the spectrum or spectra you want to exclude or assign to the other data set repeat the process Bruker Optik GmbH OPUS QUANT 89 Reference Section Prediction vs True Methanol Cross Validation 48 8 49 25 49 45 49 65 49 86 60 05 50 25 Prediction vs True Methanol Cross Validation 49 34 49 32 49 3 49 28 49 26 49 24 49 22 49 2 49 18 49 16 49 14 49 12 49 1 49 9 49 92 4995 49 98 50 50 02 50 05 50 08 50 1 Figure 77 Zooming into one particular spectrum To undo the zoom click with the right mouse button Once you have selected the spectra click on Special to open the following dialog box Transfer Spectra to another Data Set b xj Set Spectra on Excluded If you click on the first button all spectra in the zoomed graph are set on Excluded Test gt Calibration If you click on the second button all selected spectra are moved to the opposite data set Calibration gt Test or Test gt Calibration Repeat spectra spectra with the same Sample Number are also moved Figure 78 Transfer Spectra to another Data Set or Exclude Click on the Set Spectra on Excluded
99. is distance is displayed for the selected method 10 9 Setup Quant 2 Method Optimize The OPUS QUANT software facilitates the optimization of a Quant method Number RMSECV Rank Regions Preprocessing m Optimize status Figure 85 Setup Quant 2 Method Optimize Page 96 OPUS QUANT Bruker Optik GmbH Setup Quant 2 Method Optimize The optimization is done automatically by successively trying a combination of predefined frequency regions and data preprocessing methods The result of the optimization run is a list showing the Rank and RMSECV value for each combi nation of predefined frequency regions and data preprocessing methods See figure 86 Note Derivatives are smoothed using the settings defined on the Parameters page However on the basis of the optimization results you have to find out yourself which the combination yields the best result Afterwards perform a validation using these parameters and have a closer look at the results Note It is not necessary that you first validate your method alternatively you can perform an optimization An optimization run can be very time consuming optimizing a test set valida tion is usually a matter of minutes while optimizing a cross validation can take up to several hours You can stop a running optimization procedure at any time See figure 87 You also have the possibility to limit the number of preprocessing options and t
100. ise if a spectrum has been mea sured under irregular conditions 6 OPUS QUANT Bruker Optik GmbH The A values are always smaller than 1 and the total sum of all 4 is equal to the rank R Dh E 2 11 R M is the mean leverage value 5 R M is generally a suitable limit for detecting outliers If the A value is bigger than the indicated limit Factor Rank M Limit 2 12 the spectrum should possibly be removed from the list of standards Factor can range between 2 and 10 Figure 1 is an example of the distribution of the lever age values for the calibration spectra Lever Figure 1 Leverage Values Plotted against the Sample Number The sequence of the leverage values as a function of the concentration is fre quently parabolic for a one component system see Figure 2 The lowest and the highest concentration values have the largest leverage values The leverage values which are above the limit are not outliers as it might be suspected The user must be very careful in removing spectra from the calibration list for a one component system Bruker Optik GmbH OPUS QUANT 7 Theoretical Background The expression Factor Rank a 2 13 is also used as a limit for the Mahalanobis distance MahDist which is calcu lated in the analysis of unknown samples A Factor of 2 has been found to be too conservative for analysis Too many spectra are marked as outliers although the predicted compon
101. le click on the upper tile on the left side see circle in figure 50 To rearrange the order of the spectra select them click on the tile and move them to the new position while pressing the left mouse button To sort the table by a certain value e g concentration value of methanol double click on the column titles in the header 62 OPUS QUANT Bruker Optik GmbH Setup Quant 2 Method Spectra The column Data Set classifies whether a spectrum is assigned to the calibration set the test set or in case of an outlier excluded from the data set In this way you can exclude spectra from the validation without removing them from the spectrum list The default setting is Calibration In case of a test set validation this setting assigns the spectrum to the calibration set If you want to perform a cross validation all spectra have to be part of the calibration set d Excluded E Calibration 3 Figure 51 Assigning a Data Set Type Add Spectra Click on the Add Spectra button to open a Setup Quant 2 Method Select Stan dards dialog box Navigate to the target directory that contains the spectra used for the QUANT 2 method and load all spectra of interest These spectra will be added to the table Set Sample Numbers The spectra are numbered consecutively according to the order in which they have been loaded In practice you mostly acquire more than one spectrum per sample Therefore you have to adjust the nu
102. lected datapoints 0 Preprocessing No spectral data preprocessing Load Method V Load existing validation results Figure 9 Setup Quant 2 Method Load Method Page 2 This page allows you to specify the components of your sample Click on the Add Component button to create a new entry in the list The entry will be named Comp 1 You can change its name as well as the unit by selecting it in the list and editing the Name and Unit fields You can also remove entries from the list by selecting them and pressing the Delete key on your keyboard Now add three components name them methanol ethanol and propanol and enter a unit e g mg In addition you can specify the formatting of the prediction value in the Quant 2 analysis report You can choose between Default Settings 5 Significant Digits and Digits after the Dec imal Point i e you can specify the number of digits after the decimal 20 OPUS QUANT Bruker Optik GmbH point by clicking on the corresponding option button The selected for matting option has an effect on the prediction values in Quant report of the Quantitative Analysis 2 function figure 90 and the analysis results of the Quant 2 Analysis File List function figure 93 Note that the selected option applies to all components Setup Quant 2 Method New x Load Method Components Spectra Parameters Validate Graph Report Store Method Optimize Settings Add Component
103. lecting from the Rank drop down list The result for a model using only one factor of the matrix to analyze the internal test samples is shown Obviously the prediction is not very useful and the model needs to be improved Select different ranks and notice the improve ment of the prediction by either looking at the match between the predicted and the true concentration values or at the RMSECV values Browse between these ranks by placing the cursor in the Rank drop down list and using the arrow keys of keyboard If you position the cursor on one of the data points its x and y val ues as well as the sample name are displayed Prediction vs True Methanol Cross Validation 30 o5 10 20 30 4 50 EU 70 50 SU a Prediction vs True Methanol Cross Validation 10 0 510 20 30 4 50 EU ZU so SU IU Figure 22 Results for Models Employing 1 and 2 Factors 32 OPUS QUANT Bruker Optik GmbH Taking a closer Look at the Results 100 4 A e e E EO E EA ERA EO EO EE KE KE ESA A BE Ee 10 aT OO EEE EEE sss E EO BAGARE EEE BEZ 10 0 510 20 Prediction vs True Methanol Cross Validation Gia ra EEE EEE HERAF Eesi e tert b tT a EEEF EEE BEIZ EEOUUOOO See Ses He BRA EEA ER EEE He AAA AA AAA AAA ORAA AA AAA AAA 1 0 5 10 2 30 40 50 EU 70 SU 90 100 Prediction vs True Methanol Cross Validation 30 40 50 6 70 80 SU I
104. libration Multivariate calibrations make use of not only a single spectral point but take into account spectral features over a wide range Therefore the analysis of over lapping spectral bands or broad peaks becomes feasible The information con tained in the spectra of the calibration samples will be compared to the information of the concentration values using a PLS regression The method assumes that systematic variations observed in the spectra are a consequence of the concentration change of the components However the correlation between the components concentration and the change in the infrared signal does not have to be a linear one Multivariate calibrations require a large number of calibration samples and yield a large amount of data several spectra with hundreds or thousands of rel evant data points In order to conveniently handle the data the spectral data and the concentration data are written in the form of matrices where each row in the spectral data matrix represents a sample spectrum The concentration data matrix contains the corresponding concentration values of the samples The matrices will be broken down into their Eigenvectors which are called factors or principal components The advantage of this approach is that not all of the prin cipal components are necessary to describe the relevant spectral features for example some of these vectors simply represent the spectral noise of the mea surement Only the relev
105. libration Regions This table allows you to restrict the frequency region for the validation to one or more frequency region s The limits of the frequency region s can either be entered manually in the table or specified interactively by clicking on the Inter active Region Selection button Interactive Region Selection If you click on this button the Select Frequency Range s window opens dis playing the calibration spectra To specify a frequency region right click on the window and select the Add Region function from the pop up menu The selected frequency region limits are indicated by gray borders The spectral range shown on the white background will be used for the preprocessing You can move the borders by placing the cursor on them and shifting them while pressing the left mouse button Positioning the cursor on the white area allows you to shift the whole region in the same way You can also add several fre quency regions by right clicking on the window and selecting the Add Region function from the pop up menu To delete a region call up the pop up menu again and select Remove Alternatively you can delete a region from the fre quency table on the Parameters page The pop up menu also provides a zoom function and a cross hair cursor to conveniently get an exact reading of a spec tral data point For detailed information refer to the OPUS Reference Manual Select Frequency Range s 12000 11000 10000 9000 3000 7000
106. libration spectra and regions with only a few samples So this value helps you to decide whether an analyzed spectrum is use ful for your Quant 2 model or not To calculate the component value density 10 of the calibration spectra are considered Note that this value is not calculated for a Quant 2 model with less than 30 spectra 104 OPUS QUANT Bruker Optik GmbH Quant 2 Analysis File List Methods 10 12 Quant 2 Analysis File List Methods In case you have set up several Quant 2 methods for a given sample set using different parameters i e different data preprocessing methods and frequency regions the Quant 2 Analysis File List function allows you to compare these methods with each other in order to find out the most capable method for your analytical purpose This is done using independent samples i e samples that have been neither part of the calibration set nor of the test set It is highly rec ommended to use this function for a final check of the robustness of the calibra tion models you have set up You find this function in the Evaluate menu Methods Spectra Analysis Results Graph Statistics Add Methods Load Method List Save Method List Clear ur File Name Components C QUANT Quant Examples Sunflower Seeds grinded Protein 1 q2 Protein Eze C QUANT Quant Examples Sunflower Seeds grinded Protein 2 q2 Protein Ea C QUANT Quant Examples Sunflower Seeds grinded Protein 3 q2
107. lly 100 Specify the minimum and maximum concentration value for each component Note that the sum of the average values minimum maxi mum 2 must be equal to the specified sum of component values Enter the component names and the number of samples from 15 to 100 samples you want to include Then click on the Search Component Values button Bruker Optik GmbH OPUS QUANT 53 Calibration Design By default the software calculates an uniform distribution of the concentration values i e if the concentration range is divided into three subranges the number of samples in the low middle and high subrange is nearly equal If you deacti vate the corresponding check box the software searches a non uniform distribu tion of the concentration values if it is necessary e g Comp 1 0 80 Comp 2 0 90 Comp 3 0 100 Comp 4 0 10 The following figure shows an example for a four component system Calibration Design E x Setup Table Graph Number of Components 4 Number of samples range 15 100 jas Sum of component values fi 00 Search Component Values IV Uniform distribution of concentration values m Instruction Select the number of components used for the calibration design Enter the wanted sum of component values for one sample In many cases this is 100 percent Specify the minimum and maximum concentration values for each component The sum of the aver
108. lly the spectra and marks them by a red asterisks The selection is done from the aspect of covering the whole concentration range in the best pos sible way to obtain a robust model To undo the selection click on the Exit but ton Specify whether the spectra that are selected by the program are set to Excluded or assigned to the Test Data Set Then click on the Set Data Set but ton The selection is displayed in the spectra list on the Spectra page 80 OPUS QUANT Bruker Optik GmbH Setup Quant 2 Method Parameters 0 100 0 000 0 100 10000 9750 9500 9250 9000 8750 3500 8250 8000 7750 Go to Setup Quant First derivative MSC File Name Enalapril_Tabletten color_PCA_Loading_1 0 Enalapril_Tabletten color_PCA_Loading_2 0 Enalapril_Tabletten color__PCA_Loading_3 0 Enalapril Tabletten color_PCA_Loading_4 0 lt 1 lt 1 lt 1 lt 1 Soe ic Interactive Region Selection SE Figure 69 Loading Diagram The loadings factors describe the weighting of the individual x variables with regard to their contribution to the variance The loadings allow you to determine which data points make the biggest variance between the samples and to assess the importance of the individual variable for the calibration The function of the buttons and check boxes in this window are described in section Display Preprocessed Spectra Bruker Optik GmbH OPUS QUANT 81 Reference Secti
109. ls using a few test samples first Make sure to use the same parameter set during the measurements of the calibration set that you later want to use for the analysis Now that you have all spectra at your hand you should decide on whether you want to use the whole frequency region of the data and whether you want to per form some data preprocessing before starting the QUANT software Frequency Region The PLS regression method is a full spectrum method the chemometric model should improve with an increasing number of data points However in some cases spectral noise or additional components in the samples may cause the PLS algorithm to interpret these features which can degrade the model In these cases it is advisable to limit the frequency region used for the PLS regres sion Usually this step is taken to improve a regression that did not yield a satis 16 OPUS QUANT Bruker Optik GmbH Acquiring Spectra and Data Preprocessing factory model When narrowing down a spectrum to a few absorption bands it is found that in general bands between 0 7 and 1 0 absorbance units AU gener ate the best results Values greater than 2 5 should not be used Also it is not necessary to identify substance specific peaks but rather to include the com plete frequency region of the functional groups e g alcohols from a spectrum Nevertheless in case of a minor component it can be helpful to know the absorptions in the spectrum to fi
110. lso the option to print the history We advise you to select the setting Landscape with your printer Otherwise the content may not fit on the page 118 OPUS QUANT Bruker Optik GmbH 1 3 Method Protection The OPUS software allows the protection of Quant 2 and Identity Test methods A typical scenario for protecting a method is User A has set up a method and wants to give this method to user B but he does not want that user B passes the method to anybody else Therefore user A protects the method for the spec trometer of user B by indicating the MAC ID of spectrometer of user B Note A protected method can only be used if the computer on which the OPUS software is running is connected to the spectrometer with the indicated MAC ID Each spectrometer of a certain instrument series e g MATRIX TENSOR 27 TENSOR 37 MPA VERTEX has a unique MAC ID The protection is related to this MAC ID Method protection is not possible with older spectrometers that are equipped with an AQP e g Vector 22 or Equinox Protecting a Method 1 User B tells user A the MAC ID of his spectrometer User B can get the MAC ID either via the OPUS software or via the Internet Explorer To get the MAC ID via the OPUS software select in the OPUS Measure menu the Direct Command Entry function and enter the command CON FIG MCID Then click on the Send Command button The MAC ID e g 00 00 AD 07 AB 11 is displayed in the lower part of the dialog window
111. m press the Shift or Control key while selecting the spectra in question and pressing the Delete key Bruker Optik GmbH OPUS QUANT 71 Reference Section 10 4 Setup Quant 2 Method Parameters Setup Quant 2 Method C OPUS 6 0 Data Extended Demodata QuantTutorial Alk_d2 q2 Load Method Components Spectra Parameters Validate Graph Report Store Method Optimize Settings CIV Preprocessing in individual regions PS Set m Preprocessing in calibration regions No spectral data preprocessing DI je Mean Centering 1 r Calibration regions Interactive Region Selection Clear Selected Regions View spectra M PCA Display Preprocessed Spectra __Display Preprocessed Specta Factors b Factorize Show every x th sample x x j3 Show Scores Show Loadings Sample Statistics Figure 61 Setup Quant 2 Method Parameters Page Preprocessing in individual Regions PS When you activate the Preprocessing in individual regions PS check box and click on the Set button the following dialog window opens Preprocessing in individual regions Sequence xj Do not use a Quant 2 method with this option in a previous OPUS version 5 5 and lower Vector normalization SNV z am o Interactive Region Selection hh BR Add to List Preprocessing sequence P5 Figure 62 Data Preprocessing
112. m runs separately for each component you have indi cated before As soon as the calculation is finished the result will auto matically be displayed by switching to the Graph page 28 OPUS QUANT Bruker Optik GmbH Performing the Validation Setup Quant 2 Method C OPUS 6 0 Data Extended Demodata QuantTutorial Alk_ x Load Method Components Spectra Parameters Validate Graph Report Store Method Optimize Settings Validation Parameters 1 Methanol 10 M Ethanol 10 b No of samples leaving out fi Propanol 10 b Validate m Calculation Status Methanol Cross Validation EE Figure 18 Setup Quant 2 Method Validation in Progress 7 Figure 19 shows the diagrammatic representation of the validation result By default the predicted concentration values versus the true concentration values i e the concentration values you have entered on the Spectra page are displayed Outliers are marked in red The recom mended rank Rec is in our case 6 The results of the predicted concen tration values are displayed for this rank but the display can be changed by selecting a different rank in the Rank drop down list In addition the name of the validation the component for which the result is shown as well as the values for RMSECV root mean square error of cross valida tion and R coefficient of determination are displayed For detailed information refer to cha
113. mbering in the Sample column accordingly Click on the Set Sample Numbers button and specify how many spectra belong to one sample see figure 52 However this requires that the spectra of the same sample are listed in groups in the table i e all spectra of sample one followed by all the spectra of sample two etc Note Take into consideration that a correct sample numbering i e the correct assignment of several spectra that belong to one sample is of crucial importance for the validation If you have acquired different numbers of spectra per sample you have to apply different sample number settings on respective parts of the spectra table To do this select only the first and the last row e g row 9 and 17 of that part of the spectra table to which you want to assign a different number of spectra per sample than to the rest of the spectra table by selecting the first spec trum of the part in question pressing the CTRL key and clicking on the last spec trum of the part of the spectra table To facilitate the sample number setting activate the Set sample numbers according to file names check box In this case files with the same file name but a different extension are assigned to the same sample number Bruker Optik GmbH OPUS QUANT 63 Reference Section Set Sample Numbers x Number of spectra per sample ET I Set sample numbers according to file names Lea Figure 52 Set Sample Numbers Dialog Bo
114. mple Spectrum vs Sample Number 36 0 00193337 lo 108 Mva4616k 0 N T Q I ARORA T T T j T T l ARGA A AKE 15 17 19 21 23 2 2 2 3 33 a 39 41 43 Figure 65 Sample Statistics Plot This plot shows the difference RMS root mean square between each spec trum of a sample and the corresponding calculated mean spectrum versus the sample number It allows you to recognize outliers that occurred when measur ing the same sample repeatedly and exclude them in the spectra list in the run up to validation Note The evaluation is based on the selected frequency regions and the prepro cessed spectra When you place the cursor in the plot on a data point a tooltip appears contain ing the following information sample number spectrum number file name and the exact difference value See figure 65 Note The sample statistics plot can only be displayed if the data set contains sev eral spectra of each sample Otherwise a corresponding message window appears PCA The purpose of the PCA is e selecting optimally spectra for the test data set as an alternative to the procedure described in section 11 3 subsection Set Data Set e selecting suitable spectra for calibration set e getting an overview of the acquired spectra and e recognizing outliers When you activate the
115. ms the Bruker spectrometer and the foreign spectrometer Use exactly the same samples for both spectrometer systems not different samples from the same material Ideally the samples should cover the complete range of the parameters to be analyzed Moreover it is of crucial importance that the mea surements with both spectrometers are performed under the same environmental conditions and without delay These measures ensure that the differences between the foreign spectrum and the OPUS spectrum of one and the same sample reflect only the different characteristics of the two spectrometers Note To be able to import the data into OPUS the measured spectra need to stored in a 3D JCAMP multifile For information about how to create a 3D JCAMP multifile refer to the corresponding software manual 14 1 Setting up a Spectra Transfer Method Converting the original Data In case the forgein spectra are not available in the OPUS format they have to be converted to this format before the spectra transfer Moreover if the x axis of the original spectra has another unit e g nm or um than the OPUS spectra cm wave number the x axis unit of the original spectra has to be converted into cm To do this select in the Evaluate menu the Setup Spectra Transfer Method function and click on the Convert 3D JCAMP File to OPUS Files tab See figure 114 Bruker Optik GmbH OPUS QUANT 125 Spectra Transfer Setup Spectra Transfer Method
116. nd relevant frequency regions Data Preprocessing Data preprocessing is an important stage in performing a calibration To ensure the reproducibility of the calibration samples several spectra of each sample have to be acquired If the spectra of the same sample are not identical a data preprocessing procedure must be chosen to bring them into line with each other Data preprocessing can eliminate variations in offset or different linear base lines In quantitative analysis it is assumed that the layer thickness i e the effective pathlength of the infrared light in the sample is identical in all measurements A lack of reproducibility in sample preparation can easily cause variations in sample thickness If the thicknesses are different or unknown this effect can be eliminated by a normalization of the spectra The purpose of data preprocessing is to ensure a good correlation between the spectral data and the concentration values The following methods can be applied e Linear Offset Subtraction shifts the spectra in order to set the y min imum to zero e Straight Line Subtraction fits a straight line to the spectrum and sub tracts it This accounts for a tilt in the recorded spectrum e Vector Normalization normalizes a spectrum by first calculating the average intensity value and subsequent subtraction of this value from the spectrum Then the sum of the squared intensities is calculated and the spectrum is divided by the squar
117. ne subregion After the best subregion has been found a second subregion is added After the best combination of two subregions has been found a third sub region is added and so on The best combination of subregions is searched by adding and leaving out further subregions Note For detailed information on how to specify the frequency region for the optimization also refer to section User defined Optimization Regions below Note Both optimization types depend on the selected frequency region The results of both types may be different Find the best optimization type by trial and error Note You can start several optimizations at once These optimizations are then processed one after the other e g overnight To do this click on the Settings tab and activate the Run Optimization in Background check box Open several QUANT windows parallel load the corresponding methods and select the desired parameters Then start the first optimization As soon as the white percentage progress bar appears start the remaining optimizations When you click on the Optimize button in the other QUANT windows the percentage in the bar flashes once and the QUANT window remains open As soon as the first optimization has been completed the second optimization is being processed and so on After all optimizations have been completed all QUANT windows are open again and the result lists are displayed Optimize Start the optimization by clicking the Optimize
118. nent increases proportionally In case of sample mixture containing three or more components a collinearity between two components can often not be detected on first sight The Calibration Design function helps you to find the optimal concentration distribution of sample components for a given number of samples and to avoid collinearity This function yields independent concentration values calculated at random Select in the OPUS Evaluate menu the Calibration Design function The fol lowing dialog box Calibration Design xj Setup Table Graph Number of Components SI X Number of samples range 15 100 jas Sum of component values 100 Search Component Values IV Uniform distribution of concentration values r Instruction Select the number of components used for the calibration design Enter the wanted sum of component values for one sample In many cases this is 100 percent Specify the minimum and maximum concentration values for each component The sum of the average values minimum maximum 2 must be equal to the wanted sum of component values Enter names for the components the number of samples you need and click on the button Search Component Values Figure 41 Quant 2 Calibration Design Setup Follow the instructions given in the right side of the dialog box Select the num ber of components Enter the wanted sum of the component values for one sam ple norma
119. nt Click on this button to print out the result table Enter a title for the print in the Print Title field Window The Window button switches to an OPUS report window listing the analysis results Spectral Residuals If you activate this check box in addition the spectral residuals FValue and FProb value are listed in the table Bruker Optik GmbH OPUS QUANT 107 Reference Section Quant 2 Analysis File List xj 10 15 Quant 2 Analysis File List Graph Methods Spectra Analysis Results Graph Statistics Sunflower Seeds grinded Protein 1 q2 Protein E Save Print Prediction vs True Protein Sunflower Seeds grinded Protein 1 q2 184 123 127 131 135 139 143 147 151 165 159 163 167 171 175 179 RMSEP 0 283 Bias 0 085 RPD 5 54 Offset 0 151 Slope 0 985 Corr Coeff 0 9837 Figure 94 Simultaneous Evaluation of several Quant 2 Methods Graph Page This page allows the display of the following three plot types Prediction vs No i e sample number Prediction vs True and Difference vs True These plot types can be displayed for each Quant 2 method you have used for the analysis Note If no true component values are entered in the spectra list on the previous page only the first plot type Prediction vs No is included in the drop down list The display of the
120. o meet your requirements They are specified when the signature is set up Setup Signature Setup and should generally describe the purpose of the sig nature in this example the approval of the method For further information refer to the OPUS VALIDATION Manual Click on the Sign button The signature is added to the list stating the First Name and the Last Name of the signer as well as the Meaning the Category the Date and the Time of the signature Bruker Optik GmbH OPUS QUANT 117 21 CFR part 11 Compliance Methods Add Signature Show History C OPUS 6 0 Data Extended Demodata Quant Tutoriali x Signature History Load Method Quant 2 Method q2 DI Add Signature H CFR 11 Print Signature BRI EO meaning Category Dae ime 4 Marion Fechner Review Approved Review 2004 01 20 11 54 35 2 Marion Fechner Release Approved Release 2004 01 20 11 56 31 Figure 102 Signature Added to the List You can print out a hardcopy of the signature by clicking on Print Signature If you click on the History tab you can see the history the method file Methods Add Signature Show History C OPUS 6 0 Data Extended Demodata QuantTutoriali xj Signature History Print History use Landscape 1 2 3 Add Signature Signature 2004 01 20 11 54 35 GMT 1 Signed by Mario 5 add Signature Signature 2004 01 20 11 56 31 GMT 1 Signed by Mario Figure 103 Method History There is a
121. on 10 5 Setup Quant 2 Method Validate Setup Quant 2 Method C OPUS 6 0 Data Extended Demodata QuantTutorial Alk_ x m Validation parameters Component __ Max Rank Methanol 10 Ethanol 10 Propanol 10 Load Method Components Spectra Parameters Validate Graph Report Store Method Optimize Settings Use Cross Validation E b No of samples leaving out fi GOI Validate r Calculation status Figure 70 Setup Quant 2 Method Validate Page Validation Parameters By default this table lists all components you have defined on the Components page However only those components are included in the validation process which have a check mark in the Use field The Max Rank column allows you to restrict the validation only to be calculated up to a certain rank Validation Type From the drop down list you can choose between cross validation and test set validation In order to perform a test set validation you have to define a test set on the Spectra page first No of Samples Leaving Out In case of a cross validation you must specify the number of samples to be left out and used to test the validation cycle Note If you have acquired several spectra from one sample ensure that all spec tra of one sample are assigned to the same sample number 82 OPUS QUANT Bruker Optik GmbH Setup Quant 2 Method Validate V
122. on for the analyzed system can be found in the this example from 8800 to 4500 cml The lower part of the window shows a list containing the b vector and the PLS vectors In the Show column you can select the files to be displayed To view the PLS vectors deactivate the check box of the b vector RegrCoeff and re scale by right clicking and selecting Scale all Spectra erg Show Everything XY from the pop up menu The higher the number of the PLS vector is the more noise is visible The fol lowing figure shows a comparison between the first and the sixth PLS vector of the given example PLS V6 0 000 0 050 PLS V1 0 000 5800 5600 5400 5200 5000 4800 4600 4400 4200 4000 Figure 74 Comparison of two PLS vectors The Go to Setup QUANT button brings you back to the previous page The Interactive Region Selection button has already been described above Special Click on this button if you want to exclude spectra from the validation or assign spectra of the calibration set to the test set or vice versa Beforehand you have to select the spectra Zoom into the area of interest by dragging a box around the spectra while pressing the left mouse button 88 OPUS QUANT Bruker Optik GmbH Setup Quant 2 Method Graph Prediction vs True Methanol TI Cross Validation Figure 75 Zooming into a group of spectra The area will be enlarged Predict
123. on the Set button and then on Exit button Set Sample Numbers x Number of spectra per sample jd J Set sample numbers according to file names Dea Figure 12 Set Sample Numbers Window 22 OPUS QUANT Bruker Optik GmbH As a result now two spectra are assigned to one sample Setup Quant 2 Method C OPUS 6 0 Data Extended Demodata QuantTutorial Alk_d2 q2 Load Method Components Spectra Parameters Validate Graph Report Store Method Optimize Settings Add Spectra Change Path Copy Spectra Window Set Sample Numbers Set Data Set Comp Correlations Print E re E EEE 4 Calibration 1 C OPUS 6 0 Data Ext O54LK1 1 2 Calibration 1 CAOPUS 6 0 Data Ext O54LK1 2 3 Calibration 2 CAOPUS 6 0 Data Ext O54LK2 1 4 Calibration 2 CAOPUS 6 0 Data Ext O54LK2 2 5 Calibration 3 CAOPUS 6 0 Data Ext O54LK3 1 6 Calibration 3 CAOPUS 6 0DataExt 05ALK3 2 Calibration 4 CAOPUS 6 0 Data Ext O54LK4 1 GS Calibration 4 CAOPUS 6 0 Data Ext O54LK4 2 a Calibration 5 CAOPUS 6 0 DatalExt 05A4LK5 1 140 Calibration 5 CAOPUS 6 0 Data Ext O54LK5 2 141 Calibration 6 CAOPUS 6 0 Data Ext O54LK6 1 142 Calibration 6 CAOPUS 6 0 Data Ext O54LK6 2 a3 Calibration V C AOPUS 6 0DatatExt O5ALK7 1 44 Calibration 7 CAOPUS 6 0 Data Ext O5ALK7 2 45 Calibration 8 CAOPUS 6 0 Data Ext O54LK8 1 46 Calibration 8 C AOPUS 6 0
124. onsist of about the same number of samples and each set should cover the whole concen tration range of your system Needless to say that a sample must not be included in both sets The advantage of the test set method is the speed of calculation when dealing with a very large number of samples Sometimes this method is even required e g for governmental regulations Calibration Set Test Sample Developing a Method Validating the Method Figure 6 Test Set Validation 3 1 Choosing Calibration Samples The first step of building a chemometric model is to pick a sufficiently large number of samples to represent your system These samples have to be quantita tively analyzed by a reliable method to determine their components Then the IR spectra of all samples are taken and depending on the type of validation method used a calibration set and a test set is formed of these spectra 14 OPUS QUANT Bruker Optik GmbH Choosing Calibration Samples The following rules should be observed when forming a calibration set Note For setting up a calibration model using OPUS QUANT you can use up to No general recommendation can be given concerning the number of samples in a calibration set As a rule of thumb for a one component system a minimum of 20 samples should be measured Multicompo nent systems require a larger number of calibration samples 60000 spectra maximum Choose your calibration samples in a way they cover
125. orming a spectra transfer 1 Click in the Setup Spectra Transfer Method window on the Store Method page figure 120 on the Add Spectra button and select the orig inal spectra you want to transfer to OPUS The spectra are added to the table below Now click on the Transfer Spectra button As a result of this the transferred spectra are stored in the subfolder Transfer that has been created automatically in the directory of the original spectra 2 Load the original spectra you want to transfer to OPUS Select in the Evaluate menu the Transfer Spectra function Drag and drop the spectra into the Files to transfer field and load an already existing spectra trans fer method by clicking on the corresponding button See figure 121 You can perform the spectra transfer also using the already loaded method i e the method that has been used last To start the spectra transfer click on the Transfer button The transferred spectra are stored in the same directory as the original spectra In this case the original spectra are overwritten by the transferred spectra Transfer Spectra x Select Files m Files to transfer da ae C Transfer CalibrationSpectral ABU 01 02 Aa ae C Transfer CalibrationSpectral ABU 04 01 AA ae C Transfer CalibrationSpectral ABU 04 01 e ee ID Trancfort slihystinnSnectral ARLI NANI m Loaded Spectra Transfer method No Spectra Transfer method loaded Load Spectra Transfer Metho
126. ou want the results from the validation set or the calibration set being displayed by clicking on the corresponding option button Validation Selection Select the validation run of which the results you want to be displayed in the corresponding drop down list Component Choose the component in case you have performed the validation for several components of which the results you want to be displayed Graph Type Depending on whether you have activated the Validation or Calibration option button the following plots for displaying the results are available 84 OPUS QUANT Bruker Optik GmbH Setup Quant 2 Method Graph Prediction True Validation Plots the values predicted by the model versus the true component values i e the ones you have determined by a reference method Fit True Calibration Plots the values fitted by the model versus the true component values i e the ones you have determined by a reference method Difference True Validation Plots the difference between the predicted component value and the true com ponent value versus the true values i e the ones you have determined by the reference method Residuum True Calibration Plots the difference between the fitted component values and the true compo nent values the ones you have determined by a reference method RMSEP Rank or RMSECV Rank Validation Plots the RMSEP or RMSECV values versus the rank This type of graph is use
127. our method is set up and saved you must sign it with the category Release Otherwise no analysis is possible using this method To sign a method choose Methods Add Signature Show History from the OPUS Valida tion menu Methods Add Signature Show History x Signature History Load Method Quant 2 Method q2 DI Add Signature H CFR 11 Print Signature Figure 98 Add Signature Signature Page Click on the Load Method button and select the wanted method from the Sign Method dialog box The name of the method file will appear in the title bar Methods Add Signature Show History C OPUS 6 0 Data Extended Demodatai Quant d Signature History Load Method Quant 2 Method q2 DI Add Signature l Method File has no Signature Report 71 CFR I boze Figure 99 Add Signature Method Loaded 116 OPUS QUANT Bruker Optik GmbH Signing Methods Click on the Add Signature button A dialog box figure 100 appears prompt ing you to enter your User ID and Password Login for Signature 3 x GIRO User ID MAFE Password je Cancel Figure 100 Login for Signature After entering the data and clicking on the OK button the following dialog box appears x 21 CER 11 First Name Marion Last Name Fechner Meaning of Signature Figure 101 Select Meaning of Signature Now select the Meaning of the Signature These meanings can be user defined t
128. owing display win dow opens Bruker Optik GmbH OPUS QUANT 75 Reference Section 12000 11000 10000 5000 4000 Go to Setup Quant 1 Bw OSALKT 1 1 E O5ALK1 2 o2 m 05AK1 No Spectral Data Preprocessing 2 E TEAK 3 _ O5ALK3 1 3 E O5ALK3 2 E Interactive Region Selection Figure 64 Display Preprocessed Spectra The upper part of the window shows the spectra All spectra are displayed the test spectra as well as the calibration spectra Spectra which belong to the same sample number have the same color The default display limits of the window are determined by the selected frequency regions on the Parameters page The current display limits can be changed by right clicking and selecting Properties Display Limits from the pop up menu The table in the lower part of the window contains all spectra and comprises three columns The left column lists the sample numbers and the right column the spectrum filenames In the Show column you can select the spectra you want to display in the graph by activating the corresponding check box The Go to Setup Quant button brings you back to the Parameter page The Interactive Region Section button has already been described above Sample Statistics When you click on the Sample Statistics button figure 61 the following win dow opens 76 OPUS QUANT Bruker Optik GmbH Setup Quant 2 Method Parameters Difference between Spectrum and mean Sa
129. pears showing the concentration dis tribution for each sample component pair Component Correlations 0 2499 Ethanol Methanol hi Ethanol vs Methanol Gz 0 2499 EE o a 4 d d 0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 105 Figure 58 Setup Quant 2 Method Collinearity Check Bruker Optik GmbH OPUS QUANT 69 Reference Section p HI Ca For Help press F1 No Active Task In figure 58 the concentration values are evenly distributed and no collinear cor relation can be observed The R value squared correlation coefficient is below 0 7 the threshold for correlation If this value is above 0 7 the following warning will be displayed A High Correlation In this case a review of the prepared samples will be necessary Use the OPUS function Calibration Design in the Evaluate menu to find the best component concentration values for a sample set beforehand See chapter 9 Window Clicking on the Window button embeds the QUANT setup assistant in an OPUS window A dialog box comprising three buttons Spectra Graph and Report enables you to return directly to the Spectra Graph or Report page of the Setup Quant 2 Method dialog window Quant Report full_access ows 2 Operator Default Adminis laj x ZO Eile Edit View Window Measure Manipulate Evaluate Display Print Macro Validation Se
130. pectra M PEA Display Preprocessed Spectra Factors 5 Factorize Show every x th sample DI x 3 Show Scores Show Loadings Sample Statistics Figure 25 Setup Quant 2 Method Choosing Frequency Region Limits Bruker Optik GmbH OPUS QUANT 35 Improving the Model 3 4 A separate window opens displaying the spectra You can add a new fre quency region by right clicking on the window and selecting the Add Region function The frequency region marked by a white background will be used for the calculation You can move the borders of the selected frequency region by positioning the cursor on them and sliding them while pressing the left mouse button Select Frequency Range s 12000 11000 10000 9000 8000 7000 6000 5000 4000 Cancel Select Frequency Range s 3 00 2 00 Zoom JA Scale all Spectra gt Shift Curve gt Crosshair gt dd Annotation Copy Copy All Paste 0 00 Properties 12000 11000 10000 9000 8000 7000 6000 5000 4000 Cancel Figure 26 Interactive Frequency Range Selection In this example use only one continuous frequency region Set the max imum wave number to 9000 cm and the minimum wave number to 5300 eri After clicking on the OK button the interactively defined fre quency region s are added to the table 36 OPUS QUANT Bruker Optik GmbH 5 Select the option No Spectral Data Preprocessin
131. pter 2 8 Select the option RMSECV Rank from the drop down list to get a dia gram displaying the RMSECV versus the rank The value of the recom mended rank is indicated in a different color Bruker Optik GmbH OPUS QUANT 29 Validating the Model Setup Quant 2 Method C OPUS 6 0 Data Extended Demodata QuantTutorial Alk_d2 q2 x Load Method Components Spectra Parameters Validate Graph Report Store Method Optimize Settings Validation Validation No1 7 ne Calibration Prediction True M Methanol Rank fe DI Rec 6 Re 99 95 RMSECY 0 636 RPD 45 3 Bias 0 00221 b Line E Color Offset 0 122 Slope 0 996 Corr Coeff 0 9998 Setup Quant 2 Method C OPUS SA Data Extended Demodata QuantTutorial Alk_d2 q2 Load Method Components Spectra Parameters Validate Graph Report Store Method Optimize Settings Validation Validation Nol aa C Calibration RMSECY Rank Methanol zl Rank fe 7 Rec 6 Fe 99 95 12 RMSECY A 0 636 10 9 RPD 8 45 3 7 6 Bias 5 0 00221 4 3 3 Loadings 1 D I Color Figure 20 Setup Quant 2 Method Display of RMSECV against the Rank 30 OPUS QUANT Bruker Optik GmbH Taking a closer Look at the Results 5 2 Taking a closer Look at the Results The validation h
132. rameters Validate Graph Report Store Method Optimize Settings m Parameters for Prediction Validation No 1 E Cross Validation Store Method Store validation results JV Hl a E EAE E a 1 Methanol 6 99 95 0 636 E 2 Ethanol 6 99 58 1 87 E Propanol 6 6 99 79 1 31 E E Spectral residuals Factor for Mah Dist limit 6 No spectral data preprocessing Era eee Selected datapoints 2076 Standards total 30 Calibration spectra 30 Test spectra 0 Figure 83 Setup Quant 2 Method Store Method Page 94 OPUS QUANT Bruker Optik GmbH Setup Quant 2 Method Store Method On this page also additional information about the selected method is displayed As shown in figure 83 the information includes the validation type in our example Cross Validation the number of data points the number of standard samples and the number of test and calibration spectra Validation The drop down list includes all validations you have performed during your QUANT session Select the validation you want to review If you select the option Store only Spectra List Parameters a QUANT method is stored that does not contain calibration information it cannot be used in the QUANT 2 analysis This option enables you to store a method without performing a vali dation beforehand Store Validation Results If you activate this check box the validation results are saved in ad
133. rder to fulfill the 21 CFR part 11 requirements chapter 12 how to protect a method chapter 13 and how to transfer spectra that have been acquired with a different spectrometer system chapter 14 Bruker Optik GmbH OPUS QUANT 1 OPUS QUANT Bruker Optik GmbH Introduction to multivariate Calibration This introduction is intended to familiarize you with the concept of the multi variate calibration analysis on which the QUANT software is based The OPUS QUANT software package is designed for the quantitative analysis of spectra consisting of bands showing considerable overlap Usually they origi nate from samples containing one or several components in a matrix The soft ware allows to determine the concentration of more than one component in each sample simultaneously For this purpose QUANT uses a partial least square PLS fit method The purpose of calibration techniques is to correlate measured quantities like the absorption of infrared radiation with properties of the system for example the concentration of one component in a multicomponent system Usually two steps are required the calibration of the method and the analysis to determine a value of an unknown sample Let us first take a look at the univariate calibration analysis a method well known in analytical laboratory work For calibrating the system a set of calibra tion samples needs to be measured The concentration of the substance in ques tion conta
134. red calibration spectra can not be changed 6 Click on the Set button to exit the Prepare Method Protection dialog box and to return to the Methods Add Signature Show History dialog box figure 108 Now the MAC ID you have entered is displayed to the right of the Protect Method button 7 Click on the Add Signature button and sign the method file See section 12 2 Select the signature meaning Review or Release In this case do not select the signature meaning Release and Lock If the method has been protected successfully the following OPUS mes sage appears opus xl Figure 112 OPUS Message Successful Method Protection Bruker Optik GmbH OPUS QUANT 123 Method Protection 8 Close the Method Add Signature Show History dialog window by clicking on the x button in the upper right corner of the window Note Send only the protected method file with the extension q2 e g exem_1 q2 to user B Do not send the file with the extension q2v e g exem_1 q2v which contains amongst other information the validation results This file could by misused by user B Take into consideration that a protected method can only be used if the spectrometer with the corresponding MAC ID is connected Otherwise the following OPUS message appears Ce Figure 113 OPUS Message Wrong MAC ID If you want to use a protected method wait a few seconds after starting the OPUS software until the connection between the spe
135. result is automatically stored in form of a report block with the spectrum of the ana lyzed file Clicking on this block opens a report window On the left side of the report window is directory tree showing the path and the name of the file Click on the plus sign to expand the tree If you have performed several analyses with the same spectrum using e g different QUANT methods all results will be stored in the same report block They are all listed in this directory tree Bruker Optik GmbH OPUS QUANT 103 Reference Section ba Report Display full_access ows 2 Operator Default Administrator ZOO File Edit view Window Measure Manipulate Evaluate Display Print Macro Validation Setup Help PREG ZEORREK eer eee ey When selecting one of the methods with the mouse the results are displayed in the report windows on the right The upper window displays the block header and contains information about the method used In the lower window you find the predicted concentration of each component the unit the Mahalanobis dis tance Mah Dis the threshold value Limit to classify results as outliers and the component value density OPUS Browser EI Display full_access ows 1 Operator Dea TZ E C QUANT Ice spectra MAS42980 0 1 PLS Analysis Report eismwfett q2 Signed by Marion Fechner 2004 02 05 15 39 11 GMT 1 with Relea Method File eismwfett q2 2004 02 05 15 39 11 GMT 1 PE MAS42980 0 1 bazle
136. ro Validation Setup Help PRPS ZE OS BU SESH Semmes eet He BSP dda ey GIE ZZ OPUS Browser Sb Display full_access ows 1 EE MAS42980 0 1 ha OE aueri HISTORI Eg MAS42981 0 1 de ae quant fiston Eg MAsS42988 0 1 Ax ae ouant sron Eg MAsS42989 0 1 Ax ae ouant berezi MAS42990 0 1 AE QUANT HISTORY eTA AB QUANT berezi et HSS Arne QUANT berei MAS 42997 0 1 Jone ouant hersi E MAS43010 0 1 4a ae ouant sron EE MAS43011 0 1 Aa ae ouant berei STAI Jase ouant beree ei MCA42992 0 1 Ax ae ouant berei E ASEEN ja ae ouant sron E MCR40691 0 1 Aa ae ouant isror i MCR40692 0 1 AB QUANT HISTORY i MCR40704 0 1 Jane ouant berezi MCR40705 0 1 d AB QUANT HISTORY Eg MEB31233 0 1 Ax ae evant istoni af MEB40612 0 1 E E For Help press F1 E C QUANT Ice spectra Ma542980 0 1 E Quant Report AB ISEPLS Analysis Report Spectral Residuals LX PLS Analysis Report_ eismwfett g2 Signed by Marion Fechner 2004 02 05 15 39 11 GMT 1 with Relea Operator De Method File eismwfett q2 2004 02 05 15 39 11 GMT 1 gt Component Prediction Unit Mah Dist Limit Outlier Component Value Densit Fett 12 61 Yo 0 068 0 29 13 71 za ba gt 4 r Report Display full_access ows 2 Operator Default Administrator bx No Active Task uF num RF Figure 37 Quant
137. rocessing 4 E m Optimize Status Step 1 Cross Validation E Figure 86 Setup Quant 2 Method Optimization in Progress Optimization Results List The optimization results list contains the tested subregion combinations the resulting RMSECV or RMSEP value and the optimum rank obtained by a com bination The last two columns list the frequency subregion s and the type of data preprocessing used The values are added to the list as the optimization proceeds By clicking on the first two column titles the display will be sorted according to this parameter By default the list is sorted according to the RMSECV REM SEP value Status Bar Indicates the progress of the optimization and displays the type of validation currently performed Use Parameters After you have inspected the optimization results list you can copy the best combination to the Parameters page by clicking on the Use Parameters button Bruker Optik GmbH OPUS QUANT 99 Reference Section To do this first select the respective entry in the optimization results list using the left mouse button If you now switch to the Parameters page you will find that these parameters have been copied to the respective parameter fields Aborting an Optimization Right click on the green task bar and select one of the options from the pop up menu Stop Task will halt the optimization after terminating the method cur rently running Abort Task immediately term
138. rror of prediction RMSEP RMSEP EE Diery 2 22 Bruker Optik GmbH OPUS QUANT 11 Theoretical Background To summarize the setup of a reliable PLS model is an iterative process 1 Look at the validation report to select a suitable rank 2 For this rank remove possible outliers 3 Anew determination of the optimum rank is then necessary 4 Several data preprocessing options should be tested and the selected fre quency regions should be changed 12 OPUS QUANT Bruker Optik GmbH Chemometric Models and their Validation The purpose of QUANT is the quantitative analysis of an unknown multicom ponent sample However in order to perform an analysis QUANT first has to learn about your system This means you have to develop a chemometric model using a number of calibration samples of known composition that are representative for your system The IR spectra of these samples will be used by QUANT to calculate a calibration function which essentially is the model used for the analysis of unknown samples later However the model has to be evalu ated to test its reliability of prediction validation There are two validation types Cross Validation and Test Set Validation While in the latter case two different sets of samples are used the Cross Validation uses the same set of samples for calibration and validation Cross Validation Only one set of samples representative for your multicomponent s
139. sing No Spectral Data Preprocessing No Spectral Data Preprocessing No Spectral Data Preprocessing No Spectral Data Preprocessing No Spectral Data Preprocessing No Spectral Data Preprocessing No Spectral Data Preprocessing No Spectral Data Preprocessing No Spectral Data Preprocessing a Gne D Pramracassi Figure 30 Result of the Optimization Bruker Optik GmbH OPUS QUANT 41 Improving the Model 42 OPUS QUANT Bruker Optik GmbH Generating a Report and Saving the Method After you have created a chemometric model it is expedient to document the parameters of the method The QUANT software gives you the opportunity to generate a report file that contains all important information about the Quant method 1 Click on the Report tab The results of all validation runs you have per formed so far are listed You can display the results of the individual validation runs by selecting the respective validation name in the drop down list 2 Similar to the Graph page there are several drop down lists allowing to change the rank the component and the type of result The recom mended rank is also displayed The results are listed in a table instead of being represented graphically Select the result of the last validation Validation No 3 in the True Prediction view For each spectrum of the calibration set you find next to the file name the true concentration value and the predicted one as well as the diff
140. sis 77 Print 24 45 51 87 95 108 109 110 Protection mode 122 Q QUANT setup assistant 45 70 132 133 Quantitative analysis 104 R R 6 10 25 29 70 86 93 96 112 R Rank 85 Rank 6 10 12 18 28 29 32 33 37 38 39 40 43 86 88 93 96 97 113 Regression coefficient 88 Regression line 109 110 112 113 114 Removing component 61 Removing spectra 71 Report 43 93 Report block 50 104 Report window 50 104 Residual 6 113 Residuum True 85 RMESCV 32 38 RMSECV 9 10 29 31 33 37 39 40 91 93 96 97 100 113 RMSECV Rank 29 33 37 85 RMSECV RMSEP 86 RMSEE 6 85 113 RMSEP 10 11 18 91 100 109 110 113 RMSEP Rank 85 RPD 86 93 109 110 113 Run optimization in background 99 103 S Sample number 13 21 22 62 63 64 76 77 82 86 91 94 109 Save method list 106 Save spectra list 107 Score coefficient 8 78 86 Score diagram 79 80 Score matrix 78 Scores 5 78 79 80 111 SD 113 Second derivative 17 SEP 114 Set sample numbers 63 Set test sample 66 Set test spectra 65 Signing methods 116 Signing spectra 115 Slave spectrum 127 129 Slope 93 109 110 114 Spectra transfer 125 Spectra transfer method 125 127 130 Spectral data matrix 4 111 113 Spectral residuals 8 9 94 97 108 Spectrum List 62 SSE 6 114 Store method 95 96 Store validation results 96 Straight line subtraction 17 T Test data set 77 80 91 Test sample 16 31 T
141. subregions interactively by clicking on the Jnteractive Region Selection button Run Optimization in Background When you activate this check box the optimization of the QUANT method see chapter 10 9 will run in the background and other OPUS tasks can be carried out simultaneously Method Protection Activate the Store Spectra in Quant 2 Method File check box if you want to protect the method in the mode Enlarge Method or Change Parameters See also chapter 13 102 OPUS QUANT Bruker Optik GmbH Quantitative Analysis 10 11 Quantitative Analysis Quantitative Analysis 2 x Select File s m Files for Quantitative Analysis 2 m Loaded Quantitative Analysis 2 Method C QUANTSIce method EisMwFett q2 Load Quant 2 Method Analyze Cancel Help Figure 89 Quantitative Analysis Select File s File s for Quantitative Analysis 2 Select the spectrum files of your unknown samples that will be subject to a quantitative analysis First you have to load these spectra in OPUS Then select one or more absorption blocks in the OPUS browser window and drag and drop them in the File s for Quantitative Analysis 2 field Load Quant 2 Method Click on this button to define the QUANT method you want to use for the anal ysis If a method has been loaded before it will be active by default Analyze Clicking on the Analyze button starts the QUANT analysis The analysis
142. sults are listed in a table instead of being displayed graphically Report Type The following report forms are available True Prediction Instead of a graphical display the numerical values are given in form of a table The file name the true and the predicted value as well as the difference of both values are listed 92 OPUS QUANT Bruker Optik GmbH Setup Quant 2 Method Report RMSECV The rank R the RMSECV value the bias the RPD value the offset and the slope are listed The recommended rank is indicated in blue Concentration Outlier This report type is useful to identify potential outliers The file name the FProb and FValue as well as the difference between both values are listed Outliers are marked with an asterisk in a separate column To exclude these outliers click the Exclude Outliers button Validation Report This report type provides a complete report suited for documenting your valida tion method To print this report click on the Print button If you want to copy the whole report to the clipboard mark the report by click ing on the upper left tile in the Validation Report and press Ctrl C on the key board Now you paste the content of the clipboard into another software application Setup Quant 2 Method C OPUS 6 0 Data Extended Demodata QuantTutorial Alk_ x Load Method Components Spectra Parameters Validate Graph Report Store Method Optimize Settings
143. ted with their file name and path in a table and have a consecutive number assigned to For information about how to select and sort table entries refer to section 10 3 Important Note Add only spectra acquired from independent samples i e sam ples that have NOT been used for setting up the methods which are to be ana lyzed with the function Quant 2 Analysis File List Load Spectra List Use the Load Spectra List button to open a saved spectrum file list Save Spectra List If you want to use the same spectrum files repeatedly for analysis you can save the list in a file with the extension fl Add Component Columns Clicking on this button adds additional column s for each active component of the loaded Quant 2 method to the spectra table The column name corresponds with the component name Enter the true component values into the added component column s To facil itate this procedure you can paste them from the clipboard It is not necessary that for each spectrum a true component value is entered But at least two values per component must be entered for OPUS to calculate the statistics Note The true component values of the independent samples have to be deter mined by a different analytical technique reference method 106 OPUS QUANT Bruker Optik GmbH Quant 2 Analysis File List Analysis Results 10 14 Quant 2 Analysis File List Analysis Results Quant 2 Analysis File List xi Spectra Me
144. the PLS vectors have a high FValue FProb indi cates the probability that a standard is a spectral outlier Bad calibration standards can be recognized by their true values not being predicted with the remaining spectra Using the difference values an automatic outlier detection is performed to mark the samples whose deviation from the true concentration value is particularly large and statistically significant In these cases an FValue is calculated _ M 1 Differ FValue 11 6 Differ JZ FValue AF Value d FValue FProb eio 11 7 AFValue d F Value 0 Mahalanobis distance The Mahalanobis distance serves to quantify outliers During the PLS calculation the Mahalanobis distances of each calibration spec trum is determined From these values the threshold of the Mahalanobis dis tance is derived Spectra of unknown samples can be reliably analyzed using a calibration function if their Mahalanobis distance is within this threshold Offset The offset is the y value of the regression line if x 0 PRESS Predictive Residual Error Sum of Squares This value is the sum of all squared differences between true and predicted concentration M PRESS gt Diger 11 8 R The coefficient of determination R3 gives the percentage of variance present in the true component values which is reproduced in the prediction R approaches 100 as the predicted concentration values approach the true val ues Su
145. thod Optimize Settings Load Method V Load existing validation results M General information Standards total D Calibration spectra D Test spectra D pr Components 0 Frequency ranges D Selected datapoints 0 Preprocessing No spectral data preprocessing Figure 48 Setup Quant 2 Method Load Method Page Load Method If you click on the Load Method button you can load an existing Quant 2 method Quant 2 method files have the file extension q2 Note Quant method files created with the OPUS OS 2 QUANT software can also be loaded However if you store such a method file using the OPUS QUANT software this file can not be opened with OPUS OS 2 QUANT any longer To avoid this store the modified OPUS OS 2 QUANT file under a differ ent file name Bruker Optik GmbH OPUS QUANT 59 Reference Section Load existing Validation Results If an existing Quant 2 method has already been validated and the validation results have been stored together with the method you can load these results by activating the Load existing validation Results check box Otherwise only the spectra the components and the parameters of the method will be loaded when you load the method file General Information The General Information group field displays the statistical information about the loaded Quant 2 method The information includes the number of spectra calibration and test spectra and the number of
146. thods Analysis Results Analyze Print use Landscape Window Print Title I Spectral Residuals FileName Sample Name Method MAS42980 0 Pulver Av of 3 EisMwFett q2 MAS42981 0 Pulver Av of 3 EisMwFett q2 MAS42988 0 Pulver Av of 3 EisMwFett q2 MAS42989 0 Pulver Av of 3 EisMwFett q2 MAS42990 0 Pulver Av of 3 EisMwFett q2 MAS42995 0 Pulver Av of 3 EisMwFett q2 MAS42996 0 Pulver Av of 3 EisMwFett q2 MAS42997 0 Pulver Av of 3 EisMwFett q2 MAS43010 0 Pulver Av of 3 EisMwFett q2 R MAS43011 0 Pulver Av of 3 EisMwFett q2 MAS43025 0 Pulver Av of 3 EisMwFett q2 MCA42992 0 Pulver Av of 3 EisMwFett q2 MCA42993 0 Pulver Av of 3 EisMwFett q2 MCR40691 0 Pulver Av of 3 EisMwFett q2 MCR40692 0 Pulver Av of 3 EisMwFett q2 MCR40704 0 Pulver Av of 3 EisMwFett q2 RR RK RR kk KK KK KK K Figure 93 Simultaneous Evaluation of several Quant 2 Methods Analysis Results Page Analyze Starts the QUANT analysis For each file indicated on the Spectra page a QUANT analysis deploying all methods listed on the Methods page will be per formed The results are listed in form of a table comprising the results of all spectrum files The file and sample name the method used for the analysis the component analyzed and the results prediction Mahalanobis distance outlier component value density are listed In addition the analysis result table com prises also a column with the true component values you have entered at the Spectra page Pri
147. tio of standard deviation to standard error of prediction SD RPD 2 11 14 SEP SD Standard Deviation The standard deviation is a measure of the degree to which the component values of a sample set are dispersed around the mean component value The standard deviation is the square root of the variance It is calculated as follows 2 OPO i BREA 11 15 M 1 with M being the number of spectra and y being the mean component value Bruker Optik GmbH OPUS QUANT 113 Abbreviations and Formulas The mean component value is calculated as follows True di 5 a 11 16 SEP Standard Error of Prediction The standard error of prediction bias corrected is a quantitative measure for the preciseness of a test set validation It indicates the standard deviation of all bias corrected measured values from the true value The bias corrected standard error of prediction is calculated as follows gt Differ Bias SEP M 1 11 17 SSE Sum of Squared Errors The residual Res is the difference between the true and the fitted value Thus the sum of squared errors SSE is the qua dratic summation of these values SSE gt Res 11 18 114 OPUS QUANT Bruker Optik GmbH Signing Spectra 1 2 21 CFR part 11 Compliance When using the QUANT software in combination with the OPUS VALIDA TION package several rules must be observed to set up a method and perform the quantitative an
148. tion in Si X V Leave Exclude spectra IOE Test samples in Set Test Samples Set selected spectra on Set color on page Graph for selected spectra Calibration 2 Blue F E Color Set Data Set Set Color Special Setting Exit Figure 54 Set Data Set First Test Sample This field is used to specify the beginning of the test set If you have measured 120 sample spectra for example and spectra 1 to 69 form your calibration set while the remaining spectra are to be assigned to the test set you have to enter the value 70 in the First Test Sample field Block Length Test Sample This field is used to define the number of the test set spectra Taking the above example your test set comprises 50 samples Therefore set the Block Length Test Samples to 50 or a larger value Gap Calibration Samples So far we acted on the assumption that the spectra for the calibration set and the test set are loaded as continuous blocks in the spectrum table However if the spectra forming both sets are loaded alternating in the spectrum table you can still assign the spectra to the test set by specifying the spacing in the gap The assignment shown in figure 55 was created by specifying sample 3 as the First Test Sample and setting the Block Length Test Samples and the Gap Calibra tion Samples to a value of 2 Bruker Optik GmbH OPUS QUANT 65 Reference Section E asset Des ret rete wet at
149. to perform a QUANT analysis Note that only those components will be used for the analysis of which the Use check box has been activated Select Validation Results E The results of the selected validations will be stored Select All Cancel Figure 35 Setup Quant 2 Method Saving the Method The Select Validation Results window opens Select the validation s you want to store on the disk and click on the OK button The standard Save File dialog opens Enter a file name and specify the target direc tory Now you have finished the setup of the Quant 2 method Close the Setup Quant 2 Method dialog window by clicking on the cross button in the upper right corner Bruker Optik GmbH OPUS QUANT 47 Generating a Report and Saving the Method 48 OPUS QUANT Bruker Optik GmbH Performing a quantitative Analysis Compared with setting up a Quant 2 method the quantitative analysis of unknown samples is an easy task However take into consideration that the concentration values of the samples have to be within the concentration range covered by the calibration set Before you actually start the quantitative analy sis load the spectra of your unknown samples into the OPUS browser Select the Quantitative Analysis 2 function from the OPUS Evaluate menu The Quantitative Analysis 2 dialog box figure 36 opens Drag and drop the absorp tion block of the files you want to analyze from the OPUS browser in the File
150. tup Help 8x PEPE ZEGA SR ARES Beth eet KS a oD CAOPUS EE 05ALK1 1 o CAOPUS 6 0 05ALK1 2 0 100 CAOPUS 6 01 0SALK2 1 100 0 GOES 6 01 O5ALK2 2 100 0 0 CAOPUS EI 05ALK3 1 D 100 D CAOPUS 6 01 O5ALK3 2 0 100 0 CAOPUS EI 05ALK4 1 33 364 33 278 33 356 CAOPUS 6 0 O5ALK4 2 33 364 33 278 33 356 CAOPUS 6 01 OSALK5 1 49 666 25 435 24 899 GERENTEAK EE 25 435 24 899 CAOPUS EI 05ALK6 1 24 942 24 982 50078 CAOPUS 6 01 05ALK6 2 24 942 24 982 50 078 CANOPUS EI 05ALK7 1 25 392 48 95 24 658 CAOPUS 6 0 EALO 26 392 48 95 24 658 CAOPUS EI 05ALK8 1 50 017 D 49 983 CAOPUS 6 01 O5ALK8 2 50 017 D 49 983 CAOPUS EE 05ALK9 1 66 648 33 352 0 CAOPUS 6 01 O5ALK9 2 66 645 33 352 0 CAOPUS EI 05ALK101 0 33 392 66 606 CAOPUS EI 05ALK10 2 o 33 392 66 606 GREBA GAIRI DEE 0 24 914 CAOPUS EI 05ALK11 2 75 086 D 24 914 CAOPUS EE O5ALK12 4 25 425 D 74575 CAOPUS 6 0 05ALK12 2 25 245 D 74 575 CAOPUS 6 0 05ALK131 33 394 66 606 D CAOPUS EI O5ALK13 2 33 394 66 606 D CAOPUS 6 0 O5ALK14 4 D 65 944 34 055 CAOPUS 6 01 05ALK14 2 0 65 944 34 055 CAOPUS 6 0 05ALK151 33104 33 641 33 254 CAOPUS 6 0 05ALK15 2 33104 33 641 33 254 30 Calibration au r Quant Report full_access ows 2 Operator Default Administrator px uF Num RF rei Figure 59 Quant Report Window Spectra List At the bottom of the spectrum
151. user A add his own spectra perform a new vali dation and store the modified method but without having access to the calibration spectra of user A If you select the Enlarge Method mode only new spectra can be added to the method whereas if you select the Change Parameters mode also the settings on the Parameter page can be modified The protection modes Enlarge Method and Change Parameters are only available if you have activated the Store Spectra in Quant 2 Method File check box on the Setting page of the Setup Quant 2 Method dialog box before you have stored the Quant 2 method See figure 111 and 88 m Method Protection J Store Spectra in Quant 2 Method File Use this option only if you want to protect a method in the mode Enlarge Method or Change Parameters Figure 111 Setup Quant 2 Method Dialog window Store Spectra Page If a Quant 2 method file had been stored with this option and has been protected afterwards in the mode Enlarge Method or Change Parame ters the calibration spectra stored in the q2 file can only be used to perform a new validation They can not be viewed or extracted When a protected method is loaded into the Setup Quant 2 Method dia log window the calibration spectra are shown in green in the table on the Spectra page Some functions like Display Preprocessed Spectra Copy Spectra Select Test Samples and Add Component are blocked and the component concentration values of the sto
152. value in the Quant 2 analysis report can be specified You can choose between Default Settings 5 Significant Digits and Digits after the Decimal Point i e you can specify the number of digits after the decimal point by clicking on the corresponding option button The selected formatting option has an effect on the prediction values in Quant report of the Quantitative Analysis 2 function figure 90 and the analysis results of the Quant 2 Analysis File List function figure 93 Note that the selected option applies to all components Bruker Optik GmbH OPUS QUANT 61 Reference Section 10 3 Setup Quant 2 Method Spectra Setup Quant 2 Method C OPUS 6 0 Data Extended Demodata QuantTutorial Alk_d2 q2 xj Load Method Components Spectra Parameters Validate Graph Report Store Method Optimize Settings Add Spectra Change Path Copy Spectra Window Set Sample Numbers Set Data Set Comp Correlations Print FileName methanoi Seri Propano Calibration 1 CAOPUS 6 0 Data Ext 054LK1 1 Calibration 1 C AOPUS 6 0 Data Ext ISAK 2 3 Calibration 2 C OPUS 6 0Data Ext O5ALK2 1 Calibration 2 CAOPUS 6 0 Data Ext O5ALK2 2 5 Calibration 3 CAOPUS 6 0 Data Ext 054LK3 1 6 Calibration 3 C AOPUS 6 0 Data Ext O54LK3 2 Calibration 4 CAOPUS 6 0 Data Ext ISAK a Calibration 4 C AOPUS 6 0 Data
153. x Note In case of file names containing date and time of the data acquisition this part of the file name is ignored by OPUS when setting the sample numbers according to the file name See figure 53 Sample CAQUANTISpecs Pellets bright Yello File Hame Pellets bright yellow 20050310_084152 0 Pellets bright yellow 20050310_084225 0 Pellets bright yellow 120050310_084259 0 Pellets bright yellow 120050310_084331 0 Part of the file Pellets red 20050310_083146 0 name indicat Pellets red J20050310_083209 0 Pellets red 20050310_083237 0 ing date and Pellets red J20050310_083329 0 time of the data Pellets red 20050310_083353 0 Pellets white 20050310 082741 0 acquisition Calibration 1 QU Calibration I CAQUANTISpecs Calibration 1 CAQUANT Specs Calibration A CAQUANT Specs Calibration 1 CAQUANTISpecs Calibration 2 C AQUANTSpecs Calibration 2 CAQUANTISpecs Calibration 2 CAQUANT Specs Calibration 2 CAQUANT Specs Calibration 2 CAQUANT Specs Calibration 3 CAQUANTISpecs Calibration 3 CAQUANTISpecs Calibration GZ _ CAQUANTISpecs Calibration 3 CAQUANT Specs Calibration 3 CAAQUANTSpecs Pellets white 20050310_082807 0 Pellets white 20050310_082922 0 a Pellets white 20050310_083018 0 Pellets white 20050310_083044 0 Figure 53 File Names containing Date and Time of the Data Acquisition
154. ystem is used to calibrate and validate your system Before starting the calibration one sam ple is excluded from the entity of samples This sample is used for the valida tion The remaining samples are used to calibrate the system The sample used for validating the system must not be part of the calibration set Here is an example to illustrate this point let s say you choose 100 samples of a known composition From these samples you take sample number 67 and set it aside The remaining 99 samples now make up your calibration set and you will use them to create a chemometric model After doing this you will test this model against sample 67 Then you repeat this cycle this time separating a different sample e g 17 and so on until all samples have been used for validation once QUANT reiterates this cycle starting with the first sample until all sam ples have been used for validation The advantage of cross validation is the smaller number of samples required Especially if the number of samples available is limited this method should be preferred upon the test set validation Bruker Optik GmbH OPUS QUANT 13 Chemometric Models and their Validation Calibration Set Test Sample Developing a Method Validating the Method Figure 5 Cross Validation Test Set Validation The test set validation uses two independent sets of samples one for calibrating the system and the other for validating the model Both sets should c

Download Pdf Manuals

Related Search

Related Contents

Samsung 320 W 2,1Ch Soundbar H550 Užívateľská príručka  LS-880  Hewlett Packard Enterprise ProLiant DL160  Disputa amorosa acaba em morte    爪 n。b。ru スーノ 浸ーメ ガホ ン 取扱説明書(保証書付) TMW  Cisco Module/1xGENet 16xF+ENet RJ45 f/ C2600-3600  注意 - Pioneer  

Copyright © All rights reserved. Failed to retrieve file