the user manual
Contents
1. F aA B A Functional Analysis by Association FABA 1 6 User Manual Copyright 2004 by Chang Bioscience Inc All rights reserved Table of Contents IO UC Soeria creas aie a a ES 4 Compare FABA with Yeast Two hybrid oooooccccconnccccnonncnccnonncncnnnonnncnn nan nn eee e toar aiden srira tiada tanidi era eeee aaa eeesaaaeeeesaaeeeess 4 What Can FABA Be Used Fon iranieni enaa eA NAERA EENAA nyeegdnd saucedeceha needed sherds teas decane Maeensiedagecmetaaeens 5 PABA 6 Instala aia ii A A A A 6 How Does FABA Work dois 8 A Correlation score for individual dataset cccccece renee eet EEE EEE nett ae nn ane rnane rnane nne 8 B Correlation score for Multiple dalaSets oooooooccccnnnnncncooonnnnnnnnnnnnnnnnnnnnnn nn nnn ne ne ee nn PAETAE EE AEAEE E dee cena este A Kae Sinaia 8 C Gene expression data normalization ooo eee ninio 9 De MISSING Dl A A A A art 9 E Penalty for stable and low signal genes ooooicococociccnncononnnnnononnnnnononn ona rre rre seen aneneeenaanees 9 Ex NOISG IGVOIS 2 A dias 10 Power of FABA cui a A dd 11 RealiStic expectalidS iii 14 Dd IS SES 14 FABA Searches for Gene Gene Correlations ooccccconncccccnnoccnnnnnnnnnnnnnnnnnnnnnncnnnnnnnrrnnnnnnnrnnnnnnernnnnnnnrannnnnernnnnnnrrnnnnnnnnss 15 FHow to Use FABA Vi WO 000 TA A AA A AA A AA 16 E A E SS OR 16 EI arro NO 16 C BatGh QUOY ciccioviciicocic a e e e e e A e e e aE 17 D FREGUCE NOOS a A A A mare teay ass soos tare es 18 E Re2. not in a subfolder of home data yourproject g You are all set Return to FABA Viewer and run your query Warning you cannot merge two projects into one This function will be provided in the future upgrade F Delete dataset and project Delete function for dataset and project is not provided Since uploading data is time consuming delete function is intentionally omitted to avoid accidental lose of data See instructions for Advanced Users below on how dataset and project can be deleted by manipulating FABA files To remove a dataset from a project simply set the Weight to 0 step 20 Think Again Although FABA makes it much easier to find genes with similar functions by analyzing high throughput datasets from different sources it will be a mistake to think FABA as a turnkey gene function finder There are several reasons users must be very careful in reaching conclusions First there are noises in the high throughput data For example as many as 20 of clones on certain microarrays may have mistaken identities One of FABA s strength is to reduce some of these noises by comparing data from different sources however it is impossible to eliminate all the noises Second FABA searches mostly gene gene correlations which are indirect evidences that the two genes share similar functions In the absence of any definitive evidence e g genetic interactions known function similarities it is prudent to view FABA results with a li
3. 813 YDL159W CLB3 cyclin 835 YDL179W PCLY cyclin d 1545 YEROS9AW PCL6 cyclima 1924 YGL134W PCL40 cyclin Start Pause Manager Update Info Add Data Batch Figure 6 Steps for batch queries C Batch query 9 To submit for multiple queries start FABA select a project and click on Go 10 In the new window click on the Batch button lower right corner 11 Input FABA gene ID numbers separated by spaces blank space or new line Note the program can only recognize FABA gene ID numbers To find FABA gene ID numbers please use the search function in the project window FABA gene IDs are the numbers in the first column of the search results 12 Click on Start button to start the search 13 To pause resume search click on the Pause button Please be patient since the Pause command will not immediately take effect It will wait until the query in progress is completed 14 To view the progress of batch search use the history pull down menu 15 To start a new batch search pause the current batch search if any close the batch search window and then click on the Batch button again as in step 10 17 a Ze e m O re oe ES E EB jasak S G uan ER Weooitge dapan 8 X O O a A Du an ann 20 nan OG taj FF AITE EEEE Shao ae E OA O EEEE Rebuild basal noise level Rebuild cancer UNFOLDED TUO HYBRID GENE ONTOLOGY LOCALIZATION 145C FTRL DGSC TATL 058W
4. FETS LO11W SCTL YBRL49W ARAL YALO14C SYN YBRO94 YBRO79W ECM33 RLO4W YMC2 ROG9W YLR300W EXGL YALOS59W ECML YALOGOW BDHL YBRO7IW RDH54 YBLO42C FUIL YBLLO1C ECM21 YBROGSC POLIO YBLO76C ATGO YBRO93C PHOS YAROOTC RFAL YBLO39C URA YBROJAC HMT1 YBRI25C PTC4 iron ion transpor amino acid transp multicopper ferro glycerol 3 phosph aldo keto reducta SNAP receptor act transporter activ glucan 1 3 beta g R R butanediol DWA depende uridine transpo DNA polymerase p microtubule bindi acid phosphatase dameged DNA bindi CTP synthase acti protein arginine protein phosphata carnitine 0 acety structural consti ROTC feat 035W YATL 019W FUN30 TBRO99C YALA KTPI Basal Noise Noise Level 20 v Save Close Figure 7 Using lower noise level to show most reliable correlations D Reduce noises TRANS REGULATOR UNFOLDED TRANSCRIPTION MEC1 THP TOR TWO HYBRID GENE ONTOLOGY LOCALIZATION SRP 145C FTRI 058W FETS LO65W SITL YCLO20W M YIR2140 FREL iron ion transporter activity multicopper ferroxidase iron siderochrome iron ferrioxami ferric chelste reductase acti Level 2 Save Close 16 If you have added deleted data you would need to rebuild the basal noise level before running reduce noise levels Click on Basal Noise button and then Rebuild in the popup window 17 In the window that displays query results change the noise level by using th
5. data directory datalist txt Dataset index noise txt Noise statistics A i z history txt History index Individual saved data Figure 9 FABA file structures B Backup To backup all FABA data find the directory named data within the home directory of FABA Compress the data folder using data compression software such as WinZip or Stuffit Move the compressed data file into a backup folder To backup a project first find out the project data directory see C Delete project and database and compress the directory using WinZlp or Stuffit To restore to the original data uncompress the backup data and replace the files in the original folder Please avoid uncompress into a subfolder A frequent mistake will be uncompress the data folder within the original data folder e g home data data The correct structure should be home data C Delete project and dataset a Delete a project 1 Open the project index file oroject txt in home data folder The file should look like the following Mouse 24 mous Saccharomyces sacc Yeast yeas Human Huma The odd rows are project names and the even rows below are the project folder name For example the folder for the project Saccharomyces is sacc which is located at home data sacc 2 To remove a project name from the FABA software project menu delete the project name and project folder name and save the
6. program 6 If the installer can not start please check whether you have completed the download Check the installer s file size You may need to download again or request a CD from Chang Bioscience 7 Windows NT users may experience an additional problem The installer needs JAVA to start Most computers already have JAVA JRE installed Occasionally a few computers have not installed JAVA or need an upgrade Please visit the following site to install JAVA JRE first http java com en index jsp 8 After successful installation of FABA users may start the program and view pre run results 9 Start FABA Select a project and click on Go A new window will appear 10 Select from the history pull down menu right to the Go button a previous result The results will appear in a new window 11 The demo version does not include data necessary for running your own query Please request a free data CD from Chang Bioscience by emailing to info changbioscience com The data CD contains expression data d more than 3 000 microarrays for human mouse and yeast 12 After receiving the data CD follow the instructions to import data see also page 21 13 You ll then be able to run your own query and add your own or new public data You don t need the data CD if you would like to start a new project Detailed instructions are listed elsewhere in this manual 14 Please purchase a license before the demo license runs out so your work w
7. systematic experimental errors e g improper normalization of data The other symptom is that the great majority of genes associated with your query in a dataset are known to have different functions You may remove this dataset by decreasing weight or setting the weight to 0 Weight must be a non negative number between 0 and 1 21 Save your changes and run Associate again 22 Repeat steps 16 21 if necessary 23 Save the heatmap either by printing capturing screens and paste into a drawing program e g PowerPoint or click on Save to save it as a JPEG file 24 Start over with a different query To find a target gene of a signaling pathway one can use a signal gene or a known target gene as query A known target may work better because of signaling cross talks and amplification 19 F Import Data 25 If you need to upload your own data you must obtain a license for FABA However FABA Viewer allows users to import data already uploaded to FABA Following the following steps to add a project created by Chang Bioscience Inc or your colleague 26 Uncompress FABA data sent to you if necessary 27 Start FABA Viewer Click on New Project button A new window will appear 28 The new window will show a project name New Click on the Edit button next to New 29 In the popup menu replacing New with a unique project name Click on Save 30 Open the following folder home data home is
8. the number of significant correlations a gene has it is assigned one of twenty noise levels Users can choose a noise level to remove all the genes with higher noise levels from final results 10 Power of FABA signal pathway EABA Viewer localization gene onfology Vad two hybrid k oligo array cDNA arra metabolic pathway 241 0 P31E 2y Sauay protein array promoter types sequence sumilarify N hosphorylation status genetic interaction functional domains protein comple xes Figure 2 FABA can be used to analyze many types of high throughput data Shown in Figure 3 is a FABA example for Saccharomyces Cerevisiae lt includes SGD Gene Ontology data yeast two hybrid data protein localization data and 45 sets of microarray data 900 arrays The transcription profile of an uncharacterized ORF YKLO56C is similar to those of ribosome proteins and DNA RNA binding proteins suggesting a role of this uncharacterized ORF in protein synthesis Note the heatmap displays the correlation score between 2 genes in each set of experiments The name of the query gene is shown on the top The other gene member is shown on the right Each column of the heatmap represents a set of experiments Click on a column will show the data source For example the dataset Fermentation is a collection of 12 microarray data published by Olesen et al FEMS Yeast Res 2002 Dec 2 4 563 73 Red color represents a po
9. 7 GAPO localiza lOMiioiioridin aiii iia 27 D Pro QOMAINS A AA AA A AAA 27 E GONG ONO Vio tai At aa eb Reon oe os 27 Related Produc tri A A A odos 28 A BIO TOOK GOO siii anio elo Ei east Seen yaa tees aia Nooo patea 28 B The Electronic Protocol BOOKk w c0oooiiii lid 28 Frequently asked QUESOS imc i 29 1 Does FABA have any limitations on the types of high throughput data it can analyze oooiiccconicicccocannccconnnncnconnnnnnos 29 2 Do I have to use Unigene ID as identification Can use my internal database IDP ooococioninicccccacicinonnnnccnnnnanccnnnnnn o 29 3 Do you have data for any other Species ooooococicccocococonnnnnononnnnnoronnn cnn rn EEE EE EE EEE Eee r rnaar rnane EEEE 29 4 How does FABA help me don t have any HTP data c oooooncccococccincnnccccnnnnonccnnnnnnncnnnnnnccnnnnnnccnnn nn rrnnn nn rrnnnnncrnnnnnnes 29 5 How do you determine the noise level oooooconocincccnocinnccooconnccnnronnonnncnnnnnnn rn senda ee nn nn n errar rn nn re nro nnnrrnnnnnernnnnanes 29 6 Why are SOME GENES NOISY A EE no cette cnet ete AEE EE AA AA EE EE EA AA EE rn T 29 7 Which noiselevel should UE ds i 30 8 FABA results seem to change for different versions Of FABA Why ccccccsseeeeeenneeeeee ne eeeeeeaaeeeeeeaaeeeeeeaaeeeeesaaeees 30 9 Since there are so many false correlations in each individual dataset will FABA predictions be useful at all 30 10 FABA ignores differences in array types and blindly compares micro
10. across all the datasets in one project Suggested unique IDs are Systematic Name for Saccharomyces cerevisiae and Unigene ids for species such as human and mouse The info file REQUIRES a header line to avoid mistakes in omitting the Accession Number column 2 Preparation of data a High throughput data format Data file should be tab delimited text file The first column should be the unique ID field All other columns are data fields All columns and rows must have the same length Missing data can be represented by no space empty space or NA No header line is needed b Classification data format Data file should be tab delimited text file The first column should be the unique ID field The second column is the data field Classification should be represented by an integer for each class No missing data is allowed No header line is needed c Correlation data format Data file should be tab delimited text file The first and second column should be the unique ID fields The third column is the correlation data field Correlation values should be between 1 and 1 No missing data is allowed No header line is needed 21 C Data upload a Before uploading data it is strongly suggested that users upload the info file first This will soeed up the upload of the data files To upload info file start FABA go to your project and click on Update Info button A new window will appear Follow the instructions to upload the i
11. array apples and oranges How can such an approach produce any meaningful resullS 0 cece cette nce e ee ence ee eee EEE EEE EEE Ee eee ender rr ee aaeneeeeaanees 30 11 Are you absolutely sure that array data from different platforms can be COMPALEA cccccccceeeeeetteeeeeeeneeeeeenneteees 30 12 Why are you using only simple normalization options for microarray data oooiccconinicccocancccccnnonnncnnnannncnnnanannnnnannnns 30 LICENSE REQUIFEMOENtS mersa na na env Ada 32 CUSTOMEG SUPPOM rees a E fe Medial EA dci 32 Introduction FABA stands for Functional Analysis By Association FABA is designed to discover gene function by analyzing high throughput data from different sources Instead of analyzing individual data set which frequently concentrates on one process and may be biased because of system specific experimental noises FABA integrates data from different sources thus eliminating many false positive results One of the most important sources of high throughput data is the microarray gene expression data Expression data from tens of thousands of microarray experiments are now publicly available to every researcher And the amount of expression data is accumulating at a growing rate FABA is also designed to analyze this humongous amount of expression data Instead of mining data one set at a time FABA empowers scientists with a tool to analyze tens sets of data and thousands of microarrays Compare FABA with Yeast Two hybrid May ha
12. ce of gene expression correlation by using gene expression in thousands of experiments as a reference e Find potential gene functions based on that genes with similar functions frequently show similar expression patterns e Find novel candidates of a signaling pathway e Analyze many types of high throughput genome data e Incorporate gene ontology data into the analysis of high throughput data e Verify results from other experiments such as yeast two hybrid and genetic screening to find better leads FABA 1 6 Installation 1 Check system requirements Please make sure your computer has a minimum of 64 Mb of memory and Gbs of free hard drive space FABA 1 6 demo can run on system with less memory and hard drive space but if you intent to use it frequently then installing on the best computer in your lab will save you time in the future 2 Download FABA 1 6 from www ChangBioscience com Please download the install file appropriate for your computer platform Installers are available for Windows NT and Mac OS X Please contact us if you have other systems such as Linus and Unix 3 For Mac OS X uncompress the installer file by double clicking on the install zip icon A new install icon will appear after uncompressing the file by Stuffit 4 Start installation by double click the install icon Mac OS X users please make sure your system is not in the Classical Environment 5 Follow the instructions of the install program to install the
13. chine in your lab Your computer system must have at least 64 Mb of memory and 2 Gb of free space You may experience freezes when running FABA Please avoid opening other applications to conserve resources when running FABA If you plan to add your own high throughput data e g microarray data you are advised to reserve 10 Gb of free space for each 1000 arrays to be uploaded Upgrade your memory to at least 256 Mb if you expect 2000 arrays or more Please contact us if you would like us to custom upload your data Minimum System Requirements System Windows 98 and later Mac OSXand later Others Linus Unix Sun Inquire Memory gt 64 Mb 256 Mb or greater suggested Free storage space gt 2 Gb 10 Gb or more suggested 26 Public data resources As examples we list below a few public data resources that can be used for FABA analysis This list is far from complete A Gene expression data NCBI Gene Expression Omnibus GEO http www ncbi nlm nih gov geo Data depository of gene expression data including microarray and SAGE data Stanford Microarray Database http genome www5 stanford edu Yale Microarray Database http info med yale edu microarray Yeast Microarray Global Viewer http Awww transcriptome ens fr ymgv B Protein protein interaction UCLA database of interaction proteins DIP http dip doe mbi ucla edu Biomolecular Interaction Network Database BIND http Awww blueprint org bind bind php C Pr
14. croarray data which measures the transcription levels of thousands of genes Other examples include repetitive blood pressure measurements of hundreds of knockout mice under environmental conditions such as stress and diet High throughput data is not limited to experimental data Literature data such as the frequency of gene names in thousands of publications may also be used as a rough estimate of existing knowledge of gene functions The common characteristics of data in the high throughput category are 1 the data are measurements for multiple genes 2 there are multiple measurements gt 3 3 each measurement contributes equally to the final analysis Classification data are classification based on one property of a gene One example is a gene product s cellular location Each classification must be assigned an integer e g 5 for nucleus 23 for mitochondria membrane etc Correlation data are gene gene correlation data Examples include sequence similarities genetic interactions and protein protein interactions For each gene pair the correlation must be scaled to the range of 1 to 1 15 How to Use FABA Viewer A View Data 1 Start FABA Select a project and click on Go A new window will appear 2 In the query text field type in a gene id name or keywords to search for your query Hit Enter or click on Go to search Search results will appear in the window below 3 Select a query ge
15. ding structural DNA binding structural con structural col RNA binding fatty acid elo DNA binding phosphoglycera The color representation has different meanings in FABA compared to commonly used heatmap for microarrays In FABA a red square indicates that the two genes go up or down together in the specific set of data if the data is array data it does not suggest that the two genes are both up regulated Users will have to go to that dataset to find detailed information on how the two genes are correlated they could be both up or both down and most likely both up under some conditions of the experiments and both down under other conditions In the case of non array data red color simply represents same classification or positive correlation 12 Powered By Funetional finalysis By Association w 2004 Chang Bioscience Inc Es gt E a Ea o a GE E aq Sa mo Pa Nm Ba aadb EME 2 gadana e Sak pnah Ena a SEEEN ger 0 Sie BEA gdeeen soe esse aniy Correlation dun Soke Bon radar oda ES mA H 1777 Hel oH Bee Ae CESHeSaE o i Paqcee RRR aS s Py EH H H 400055 HH e ott 2 DOE a Fal oo oe me ba sioner niece SEEEE Cneeeeee gms Hee mo o ORF I Nn A AA a E pa ADOADOOD RSS Sea eee bo 5583 dd LEE ee reed 22m aca SSB ARA g E E Ml E3 310323 ARFGEF2 ADP ribosylation facti lt H5 42712 MAX MAX protein a HS 138860 ARHGAPl Rho GTPase activating E FE H3 99987 ERCC2 excision repair cross cor q 5 195825 RBPMS RNA bi
16. e pull down menu at the bottom of the result display The noise level is scaled from 1 to 20 with 20 the highest and default 18 Users are suggested to try several noise levels A rigorous statistical test of noises in gene gene correlation will be provided in the future versions of FABA 18 SIGNALING aan fed led fed wane e e BREE ite nnpec i EEEBRS ES Data set ID 12 oat tal cd pS pl le Spero Bee RR ORE Data set name Forkhead 20 Data type High throughput v bp leran Weight O Sample size 26 tin p Added by ches Zhu at al Two yeast forkhead ganes reg ott 4 ulate the cell cycle and pseudohyphal gr Beda jowth Nature 2000 Jul 6 406 5791 90 4 pict File name zhu oo A cola Last modified Thu Aug 14 22 11 29 GMT 08 00 2003 4 gt Save n Oep ne ek VERO ORCI DRA replic SLL wt mA YALAiet ATIL P Basal Noise Noise Level 20 v Save Close Figure 8 Steps for building basal noise level and changing dataset weight E Refine Your Query 19 Carefully exam the heatmap Click on a column to read the experiment information 20 Reduce weight if the dataset contributes mostly noises One symptom of noise is that a large number of genes show correlations with your query in a dataset especially if these genes are house keeping genes For microarray data high expression house keeping genes may show the same noise pattern because of
17. evelop similar software We would be glad to collaborate or provide any help we can Customer Support Thanks for using our software If you have any questions or suggestions please don t hesitate to contact us at info changbioscience com You feedback will help us to improve the software and will be greatly appreciated We ll answer your questions as soon as we can But due to limitations of our resources priority will be given to licensed users first For non licensed users please be patient Users can also post questions to the FABA user group at http www changbioscience com forum phpBB2 32
18. fine Your QUETY iii nit ne ndbibe eerie tines erin erie nine eee 19 Fe lmportsData isos as ies faces A A A A an 20 Howto Use FABA ti AA cine tacit E AA dE 21 A Data SGlGCUON A A A TAn 21 Be Data Preparation renier aea NS 21 1 Pr paration Orinio Ml o 21 2 Preparation or Uaa aeria til ida abaadadeahatanen aaa ddr ptr ad dit 21 a High throughput data format omo ia a A a a A S 21 6 Classification data Mii ii 21 c Correlation data format s ccicec ce sieeeise a Saeed clive Moe cued ae cae ce Sued bd cin Ao sue eae lives hae See ee cle eels 21 G Data Upload ss viii A A ae alae ade eet 22 D Project OxPOMt iscccscesccciivttsciveelecctiven T e E E ar eere a ar EEE Ea Ta teenies E eiar 22 PE LE TSE LE EAE IEE A A EEE PEA ada dea E E E E E E os E I E a E E TE 22 E Delete dataset and PrOjeCt eressero eero E ee OEA AEON EEEO AA EOE 23 WINK AGAIN a A E A A a A A 23 Advanced USCIS iii di ido ARE 24 A Dala OIECIOLY TIC SITU CIUIO a A A as 24 BY BackUp sports ts A ine anne apart A eee easter iA eases 24 C Delete project and Aalaset esene ai ae e eee aa EE EE AREA EE nn nn ce EE nn cnn e ai 24 SYSTEM Requirements iii A de cae Lag Dune di chy bee evel ai deeds 26 Public data TESDUICOS iprit an sina vac enea eeraa tna sand deg a KIR r etea idos 27 A Gene expression Cala v ieesisiceccieesavens o A E E shiver sevavetsebiiventeesavesissiayers staves 27 B Protein protein interaction ss seice eaa Aa Seb Sais sade dees dentate io a a o ean Ded 2
19. ill not be interrupted 15 We hope you will enjoy the FABA software please don t hesitate to ask us if you have any questions Although priorities will be giving to answer questions of licensed users we ll attempt to answer all questions as soon as we can Any suggestions and critics will be greatly appreciated 16 Please join the FABA discussion forum to share your experiences with others To join please visit http www changbioscience com forum phpBB2 How Does FABA Work To predict a gene s function scientists analyze information from different sources FABA is based on the same simple principle but doing so quantitatively For each gene pair it calculates a correlation score in individual dataset Such individual datasets could be microarray data e g microarray data under a variety of stress conditions protein localization or sequence similarities The correlation scores for these datasets will be similarity in transcription profiles co localization and sequence similarities respectively Finally a correlation score for each pair is calculated based on the scores for all the selected datasets Genes with high correlation scores are more likely having similar functions because their shared characteristics A Correlation score for individual dataset a High throughput data The most frequent high throughput data is the gene expression data The data is a serial of measurements on different samples and should be tab deli
20. imal Nevertheless the selection of house keeping genes is quiet arbitrary and many change their expression levels during developments 16 I m still confused about how can you ignore array types and data scales FABA is based on the simple principle if two genes go up and down together in only one set of data the correlation is most likely random if they go up and down in many sets of data the correlation is most likely not random The qualitative measure used to implement the simple principle in FABA 1 6 seems to work fine But the measure itself is open to improvement 17 Are you worried about the quality of the public microarray data These public data represent the current state of the art of microarray technology Until we have a good and uniform criteria to evaluate each individual dataset users need to be aware certain individual datasets may be of poor quality On the bright side a quick visual inspection of FABA results can in many cases reveal questionable datasets 31 License Requirements FABA license specifically requires users not to modify or extend FABA codes The primary reason is that we would like to maintain a data standard such that data exchange between different groups is easy This is also the reason we started FABA project in the first place It will be counter productive if different groups are not able to compare notes because different versions of software are used Please contact us if you would like to d
21. loaded by Chang Bioscience no missing data imputation is done E Penalty for stable and low signal genes To minimize detecting gene gene correlation for the large number of quasi stable genes the gene gene correlation efficient is modified to penalize stable and low signal genes P gt p R 1 R here Ris a parameter that is small if both genes are stable or have low expression values Ris much greater than 1 if either gene shows sufficient variation across samples The R value will be 0 if both genes are constants F Noise levels Distribution of Significant Correlations rrel ations Per Dataset 1000 10000 100000 Genes Figure 1 Statistics of significant correlations in gene expression data As shown in Figure 1 a small fraction of genes have a large number of significant correlations to other genes These correlations are less informative because they may not represent biologically significant co regulations Reasons for such correlations include 1 quasistable expressions 2 array system specific noises 3 fluctuation with environmental conditions such as nutrients and stress 4 hybridization noises 5 improper normalization and 6 probes contain over represented sequence elements just to name a few The noise level for each gene is determined based on the following assumption more significant correlations for a gene less informative these correlations are thus more noisy for the gene Based on
22. metimes fail to appreciate how quickly our bodies respond to environmental changes A sight of food makes our mouth wet a single breath of cigarette relaxes us and we die in five minutes without oxygen Expression of a few genes will respond quickly to environmental changes By the way if one s body does not signal enough food intakes in a few minutes one s total beta Actin will definitely increase in a short couple of days 14 Many of the genes found to be noisy by FABA seems to be house keeping genes House keeping genes have higher expression levels thus should be less noisy The definition of noise is the noise in the correlation score Quasi stable genes are noisy in FABA because they have many un informative correlations with other quasi stable genes 15 Are house keeping genes stable genes Although whether house keeping genes are stable within a factor or 1 or 2 is not an interesting question in biology it is a heated controversy in the microarray community because a number of scientists advocate data normalization based on house keeping genes expressions If you normalize array data using house keeping genes as standards then expressions for these genes are stable because that is the assumption If a different normalization procedure is used the house keeping genes are quasi stable most of the time But one can always argue that the later normalization procedure is not opt
23. mited and in the following form Gene ID Data1 Data2 Data3 Data4 The first column is unique gene identifications e g Unigene number for individual genes Data in each column represent a separate measurement The correlation between two genes is calculated as the Pearson correlation Pi lt Xi X gt Here xX is the i th row in the data matrix The correlation score is defined as e sign o p Pol 1 bol if Ip gt pol 0 if P Po here P y denotes the 95 confidence interval for the Pearson correlation p b Classification data For classification data the correlation score is defined as 1 for the same classes and 0 for different classes c Correlation data The correlation coefficient p will be directly used as the correlation score Users are suggested to correct for the confidence intervals before uploading the data B Correlation score for multiple datasets The correlation score for multiple datasets are the sum of correlation scores for individual datasets E we here W 0 W 1 is the weight parameter for dataset j C Gene expression data normalization Normalization is done for each dataset Users may choose their preferred normalization method For data uploaded by Chang Bioscience all datasets are normalized such that the means for each column are identical within individual dataset D Missing data Missing data are ignored Users may impute missing data before uploading For data up
24. nding protein witt 5 296169 RAB4R RAB4A member RAS oncoge H5 446504 ABL1 v abl Abelson murine lew 35 111 FGF9 fibroblast growth factor 9 5 380460 ICAl islet cell autoantigen l HS 442787 ZNF148 zinc finger protein 14 HS 329502 CASP9 caspase 9 apoptosis re HS 378103 RPSS ribosomal protein 55 H5 387156 GM2A GM2 ganglioside activatol HS 409934 HLA DQB1 major histocompatibi HS 436803 VBP1 von Hippel Lindau bindin lt E A3 355533 ADK adenosine kinase E HS 443960 DDX11 DEAD H Asp Glu Ala Asp E ES 80976 MKI67 antigen identified by mor HS 504644 ATM ataxia telangiectasia muti HS 385986 UBE2B ubiquitin conjugating er HS 279609 MTCH2 mitochondrial carrier he HS 262886 INPPSD inositol polyphosphate HS 154210 EDGl1 endothelial differentiat lt A E H5 184298 CDK cyclin dependent kinase HS 436593 LMAN1 lectin mannose binding HS 280342 PRKARIA protein kinase cAMP c T p 1 1 EI are p E E E m H5 79026 MLF2 myeloid leukemia factor 2 5 446414 CD47 CD47 antigen Rh related E 5 362805 MEIS2 Meisl myeloid ecotropic Ea m a H5 351874 HLA DOA major histocompatibil El H5 139851 CAV2 caveolin 2 de Basal Noise Noise Level 20 y save Close Figure 4 Tumor suppressor p53 associated genes Known genes in the p53 pathway are marked by blue arrows and potential p53 related genes are marked by orange arrows Shown in Figure 4 is the result of querying a human gene expression datasets with the tumor supp
25. ne by clicking on it in the result window 4 Click the Associate button to find genes with similar properties Search with gane name Ou ID or accession number u cyclin History 3709 YLR172C DPHS diphthine syntha Y Delete BR A 3 A Pu e a or 3 Your query matches the following genes 59 YALO40C CLN3 cyclin dependent protein kinase intrinsi ied eS pd 375 YBRIGOW CDC28 cyclin dependent protein 1784 YDL127W PCL2 cyclin dependent protein kinase intrinsic 813 YDL159W CLB3 cyclin dependent protein kinase intrinsic 835 YDL179W PCL9 cyclin dependent protein kinase intrinsic x 1545 YERDS9W PCL cyclin dependent p kinase intrinsic Salata D 1924 YGLISAW PCL10 cyciin dependent O n kinase intrinsi w Saccharompres Y ama Figure 5 Steps for viewing FABA data B View history 5 FABA will save previous searches in a list called History 6 Start FABA Select a project and click on Go A new window will appear 7 Select from the history pull down menu right to the Go button a previous result The results will appear in a new window 8 To delete saved results click on the Delete button A new window will appear Check the results to be deleted and then click on Delete in the new window 16 FABA Batch search IAN Please input FABA gene 1D numbers cyclin O History Search with gene name keyword ID or accession D cyclin 784 YDL127W PCL2 cyclin th
26. nfo file b To upload data click on the AddData button A new window will appear Select the correct data type and input the correct sample size e g 10 if your dataset has 10 data columns excluding the id field if high throughput data Click on Upload to upload c The data upload step is time consuming since a lot of computations are done in the background A batch utility is available for uploading multiple datasets Users may run the batch uploading at night or during weekends To batch upload several files prepare a tab delimited summary file for all the files to be uploaded The first row of the summary file should be the column names Name Type Scale Weight Sample Size Added By Note File Name For the Scale field the value must be exactly one of the following Unlog Log2 or Log10 The File Name field should be the absolute pathway location of the data files e g c fabadatalyeast his2 txt A pause function is provided for batch upload To prevent corruption of info file the pause will stop uploading only if one dataset has been finished Please be patient The progress bar indicates the progress in the current file being uploaded Click on Resume to continue uploading D Project export a To export a project for sharing with colleagues open the project file project txt Don t save any changes in home data folder home is where you have installed FABA This file contains the director
27. otein localization Yeast protein localization server http bioinfo mbb yale edu genome localize D Protein domains Pfam http www sanger ac uk Software Pfam E Gene Ontology Saccharomyces Genome Database SGD http www yeastgenome org Disease genes OMIN http www ncbi nlm nih gov entrez query fcgi db OMIM 27 Related Products A BioToolKit 300 BioToolKit 300 contains a number of software tools for biologists including GodlistManager and MicroHelper These tools will be handy for preparing FABA data GodlistManager has functions needed for batch GenBank search One particularly useful option is to map accession numbers to Unigene IDs MicroHelperis useful for merging raw microarray data into a single file It can also be used for filtration normalization and data transformation B The Electronic Protocol Book Perl language is frequently used by Bioinformatics scientists A number of Perl scripts are included in the Electronic Protocol Book They can be used to prepare data for FABA uploading 28 Frequently asked questions 1 Does FABA have any limitations on the types of high throughput data it can analyze The limitations are that you need to format your data into one of the following three types e 1 High throughput data Minimum sample size for this type is 4 The data should be in matrix form Gene ID Data1 Data2 Data3 Data4 e 2 Classification Data should be in two columns the first column uni
28. project index file For example after deleting the project Saccharomyces the new project index file will look like the following Mouse mous Yeast yeas Human Huma To delete project data as well remove the data folder for the project For the project Saccharomyces trash the entire sacc folder home data sacc b Delete a dataset Note Users can remove a dataset from analysis by setting the weight for the dataset to zero see step 20 1 Find the data folder for the project as in a Delete a project 2 Open the project folder i e home data saco 3 In the project folder open the dataset index file i e home data sacc datalist txt The file has the following columns Data set IDData set name Data type Scale Weight Sample size Added by Note File name Last modified Find your data set name and under the File name column find the corresponding data file name 4 Delete the row for the dataset in datalist txt file and save 5 Find and delete the corresponding data file in the project folder i e home data sacc swis for the dataset SWI SNF 25 System Requirements FABA is a JAVA application and runs on all platforms JAVA supports Installers are tested on Windows NT and Mac OSX Installers on Linus and Unix are not tested but should work We ll be mining Gigabytes of data a data size most biologists have never managed It is strongly suggested that you install FABA on the best ma
29. que gene id and the second column classification The classification should be an integer e 3 Correlation Data in three columns the first two columns are gene ids for the pair and the third column correlation values 2 Do have to use Unigene ID as identification Can use my internal database ID Sure As long as the identification is unique for each gene 3 Do you have data for any other species We rely on public available high through data We ll appreciate if you could inform us a source of HTP data we missed If there is little HTP for a particular species there is little we can do 4 How does FABA help me I don t have any HTP data Everyone can benefit from the huge amount of public HTP data Without the help of FABA it will take you days to weeks to check individually all the datasets With FABA you can find correlated genes in minutes The rigorous and consistent statistics eliminate many false positives By comparing data from a large number of sources we further reduce the number of false positives 5 How do you determine the noise level The noise level is determined based on statistics of gene gene correlations for all the experiments Briefly a gene is considered noisy if it shows a large number of correlations to other genes These genes have a large number of uninformative correlations i e noisy in the overall gene gene correlation that FABA measures 6 Why are some genes noisy We speculate that most noi
30. ressor gene p53 A total of 65 datasets and more than 1700 experiments are searched Note many genes found to share similar gene expression profiles with p53 are known p53 related genes 13 Realistic expectations Although FABA is a powerful tool and we have tried hard to make it easy to use it is still a time consuming process to get the best result from FABA Considerable amount of time is needed to carefully collect datasets from public or private sources analyze using different queries and noise levels and examine query results But compared to other wet lab techniques such as yeast two hybrid far Western or genetic screening FABA requires days of work instead of months for most wet lab experiments In addition it is complementary to wet lab approaches because its search is not limited to only one interaction As in wet lab experiments FABA will also produce a significant number of false positives Users should have realistic expectations as an evolving technology FABA has advantages compared to other wet lab technologies but considerable software development is still needed to reduce the false prediction rate Data types FABA can be used to integrate the following high throughput data types 1 Microarray data independent of array platforms 2 Protein array data 3 Genome data such as sequence similarities functional domains promoter types etc 4 Genetic data such as genetic interactions and phenotype similarities 5 Pro
31. sitive correlation and green color represents a negative correlation 11 FABA Cerevisiae 3212 YKLO56C Fa Powered By functional finalysis By Association Copyrigth 2003 Chang Bioscience NOILAIHISNYAL WOLYTNIT dad TOANN TIM ITLOHSO 0dAH dds NOTIY INIA ONITYNSIS 450 HIOHS LYH WNIHLIT SOsSIXOd4d Y 58 OHd NOIIYIOdOdS AUYNOILYLS SSTALS NIWNANIIT YS ATHOTYO ATIAI TIAJ JI TOD LAW SON ALTA YTOHSO SLYHdS0Hd ATIA0dd AWOLY TNS Td Dedsd 2osd Tdls AINS TIMS NILIHI NOILATOAI q S8 OHd q ZrdYL daL Y ZrdYL AOL JNIJY Figure 3 Genes correlated with YKLO56C a gene with unknown function ribosome subunits suggesting its function in protein translation Correlation LO56C L072C RPSSA LO87C RPL23A YBLO27W RPL19B YBRO48W RPS11B YBRO31W RPL4A YBR191W RPL21A R249C ARO4 ALOO3W EFBL YBR1S9W RPS9B YBLO92W RPL32 structural co structural ci structural ci structural c structural col structural ci 3 deoxy 7 phos translation el structural co structural col YBLOO2W HTB2 DNA binding YBLO39C URA CTP synthase a YBLOO3C HTAZ DNA binding YBRO92C PHOS YBR1S1C RPS6B YBR1I06W PHOS8 YBRO78W ECM33 YBR121C GR31 YDR225 W HTAL YBROS4C A RPL19A YNLO31C HHT2 YFLO39C ACTI YDL130W RPP1B YCLOl11C GBP2 YCRO341W FEN1 YBROl10OW HHT1 YCRO12W PGK1 PANA YKLO56C shows strong correlation with acid phosphata structural col phosphate tra glycine tRNA 1 DNA bin
32. sy genes 1 are quasistable expressions 2 fluctuate with environmental conditions such as nutrients and stress or 3 contain over represented sequence elements 29 7 Which noise level should use There is no one level for all queries Future version will provide better statistical guidelines for choosing an appropriate noise level 8 FABA results seem to change for different versions of FABA Why There are two reasons First FABA search is not exhaustive because we need to limit the number of searches such that it can finish each query in a reasonable amount of time on personal computers Newer version may have improvement in search speed thus allowing more searches to be done Second functionally correlated genes may be stable under certain experimental conditions Gorrelations of stable genes are not strong evidences of co regulation We are still experimenting to find an appropriate contribution of stable correlations to the final correlation score 9 Since there are so many false correlations in each individual dataset will FABA predictions be useful at all The false positives will be a serious problem One goal of FABA is to identify some of the false positives and to reduce false positives by comparing correlations in different datasets FABA is not perfect but nevertheless it is an improvement over analyzing single dataset 10 FABA ignores differences in array types and blindly compares microarray apples and oranges Ho
33. tein localization and protein protein interaction data 6 Metabolic and signaling pathway data 7 Gene Ontology data 8 Literature data Data that can be accepted by FABA are not limited to the afore mentioned types As a design principle we want FABA to be flexible to analyze most if not all high throughput data We want FABA to address the central question in biology gene function We want FABA to be friendly software every biologist can use Our efforts paid off Users will find that FABA is a powerful yet easy to use tool for gene function studies 14 FABA Searches for Gene Gene Correlations Behind the scene FABA searches for gene gene correlations There are many types of gene gene correlations We classified them into three categories for computational purposes High throughput Classification and Correlation A few examples of each class are listed below 1 High throughput data Examples microarray protein array high throughput mass spectrometry 2 Classification Examples protein localization conserved domain phosphorylation status signaling pathways 3 Correlation Examples sequence similarities genetic interactions two hybrid interactions literature searches High throughput data are serial measurements for each gene For each gene there must be at least four measurements and there must be more than one gene in the dataset The measurement data must be numerical A typical high throughput data is the mi
34. ttle suspicion Third the datasets may not be ideal for answering your question For example if you would like to find target genes of a signaling pathway in breast cancer your datasets must contain significant number of breast cancer data Fourth the query might not be optimal If your datasets contain mostly microarray data query with genes whose transcription levels change little will not likely be informative Of course if the query gene is absent in most datasets little correlation information will be available for the query To answer specific questions users are strongly suggested to collect their own data selection Users are also suggested to run multiple queries and examine results carefully FABA results may still be noisy but nevertheless it is a significant improvement from experiments such as single array dataset differential display or yeast two hybrid 23 Advanced Users Only users experienced in editing and managing computer files should attempt to manipulate FABA data at the file level Before any attempt in editing or deleting FABA data please make a backup copy so the data can be restored if a mistake has been made You may loose tens of hours of work if data files are improperly changed or deleted A Data directory file structure FABA Home Directory data Data Directory project txt Project Index Individual project data directory masterGeneList txt Gene info index Individual datasets hist History
35. ve similar functions Limitations Strong bait target interactions Relevant data e g gene expression False positives False positives Can use Public data Inexpensive In a sense FABA is an in silico two hybrid technology The yeast two hybrid technology uses a bait protein to fish for target proteins that physically interact with the bait FABA on the other hand find genes that share similar properties gene expression protein localization sequence similarities etc of the query gene Bothfind candidate genes that might share similar functions of the bait query gene Yeast two hybrid can only detect proteins that have strong physical contacts FABA can detect much broad gene gene interactions but relevant data must exist For example if we want to find out downstream genes of Her2 receptor in ovarian cancers the FABA data must contain significant amount of data relevant to ovarian cancers or at least to other cancer types such as breast Both yeast two hybrid and FABA will produce a large number of false positives FABA however has one advantage it can use a growing volume of public data e g NCBI GEO database has archived gt 17 000 gene expression samples at the end of April 2004 As a result FABA is a less expensive method for finding candidate genes with similar functions What Can FABA Be Used For e Find genes that show similar expression patterns across thousands of microarray experiments e Find the significan
36. w can such an approach produce any meaningful results FABA does not look for differences between microarray apples and oranges but searches for reproducible gene gene correlations in multiple datasets 11 Are you absolutely sure that array data from different platforms can be compared Yes Even more complicated meta analysis can be achieved In one study by retaining reproducible data we analyzed together three NCI60 datasets two different Affymetrix arrays and one cDNA array signal channel only All three datasets measured gene expression of a panel of 60 cell lines used in the NCI Developmental Therapeutics Program DTP A hierarchical clustering showed that the great majority of samples clustered according to their histological origins The clustering result is a significant improvement over a single dataset 12 Why are you using only simple normalization options for microarray data Our priority is to build a rough but functional program There are many other factors need to be examined and tested Normalization is one of them but not necessarily the most significant one For gene gene correlation we calculate the Pearson correlation Poor normalization will introduce error but that error may still be smaller than the uncertainty due to small sample size 30 13 What do you mean gene expressions fluctuate with environmental conditions such as nutrients and stress Dose it mean can increase my beta Actin if eat more We so
37. where you have installed FABA Viewer 31 You will see a project txt file no need to open or save project txt and a number of folders in this folder You should see a folder name that is the first four letters of your new project name or first three letters followed by a number If you are not sure which one is the one you just created check for date of creation Or open the folder you should see only three empty files datalist txt noise txt and masterGeneList txt 32 Copy all the files from the FABA data you received into this folder When prompt replace datalist txt noise txt and masterGeneList txt Files Don t drag in the top folder All data files should be within home data yourproject not in a subfolder of home data yourproject 33 You are all set Return to FABA Viewer and run your query 20 How to Use FABA A Data Selection Carefully select a number of datasets to be analyzed Selections of dataset should base on relevance and quality Good data that are not obviously relevant may actually be very helpful because they may have unexpected correlations and they may help define a baseline noise level Bad data will contribute nothing but noises B Data Preparation 1 Preparation of info file Gene info data should be a tab delimited text file in the following format Unique ID Accession Number Gene Name Description The first column is required unique ID This unique ID field should be identical for the same genes
38. y name information for each project The odd rows contain the project names and the even rows contain directory names b Find the corresponding directory in the home data folder Use file compress software e g winzip stuffit to compress the entire folder into a single file You may send the compressed file by email or burn it onto a CD E Project imports a Uncompress FABA data sent to you if necessary b Start FABA Viewer Click on New Project button A new window will appear The new window will show a project name New Click on the Edit button next to New c In the popup menu replacing New with a unique project name Click on Save d Open the following folder home data home is where you have installed FABA Viewer e You will see a project txt file no need to open or save project txt and a number of folders in this folder You should see a folder name that is the first four letters of your new project name or first three letters followed by a 22 number If you are not sure which one is the one you just created check for date of creation Or open the folder you should see only three empty files datalist txt noise txt and masterGeneList txt f Copy all the files from the FABA data you received into this folder When prompt replace datalist txt noise txt and masterGeneList txt Files Don t drag in the top folder All data files should be within home data yourproject
Download Pdf Manuals
Related Search