Anonlinesystemforfunctionalrelationshipanalysisofgenome-widegeneproductsQiangHu,Zheng-GuoZhang*DepartmentofBiomedicalEngineeringInstituteofBasicMedicalSciences,ChineseAcademyofMedicalSciencesSchoolofBasicMedicine,PekingUnionMedicalCollegeBeijing,China*Email:AbstractThoughthefunctionalrelationshipanalysisforgeneproductsisuseful,aconvenientanduser-friendlytooltomeasurethefunctionalsimilarityforgenome-widegeneproductsinmultiplespeciesisstillnotavailable.Wecomputedthefunctionalsimilarityofgeneproductsingenomewideinhuman,mouseandratbasedonouralgorithm.Databaseandwebserviceswerebuiltbasedontheprecomputedsimilarityscores.Oursystemprovidedagroupoftoolstoretrievethefunctionalsimilarityandanalysisthefunctionalrelationshipforgeneproducts.Thewebserviceisfreelyavailableat/fsim/index.html.I.INTRODUCTIONThefunctionalsimilaritymeasurementforgeneproductsisausefulmethodtoinvestigatetheirrelationship.Oneimportantapplicationoffunctionalsimilarityanalysisistopredictandassesstheprotein-proteininteractions1,2,3.Anotherapplicationistodiscoverthepositionalcandidategenesofdiseases4.Functionalsimilarityalsocanbeusedtoclustergeneexpressiondataforfunctionalrelatedgeneshavesimilarexpressionprofiles5.Mostofmethodstomeasurefunctionalsimilarityarebasedontheannotationinformationofgeneproducts.TheGeneOntology(GO)database6providesacontrolledvocabularyoftermstoannotatethefunctionsofgeneproducts.Itiswidelyadoptedbymostofalgorithmsandtoolstomeasurethefunctionalsimilarity.Thoughmanytoolshavebeendevelopedtomeasurethefunctionalsimilarity,aconvenientanduser-friendlytooltoanalysistherelationshipofgenome-widegeneproductsisstillnotavailable.TheGOtoolswebpagecollectedalotofsoftwarebasedonthedatabase.Forexample,AmiGO7andQuickGO8provideaninterfacetosearchandbrowsetheontologyandannotationdata.Therelationshipofgeneproductscanbecomparedbyusersbutnotautomatically.GOTax9thatintegratedtheannotationdataofproteinandproteinfamiliesprovidedafunctionalsimilaritysearchtool(FSST)basedonthealgorithmofInformationContent(IC)ofGOterms.Thetoolcanbeusedtomeasurethefunctionalsimilarityofproteinsandproteinfamilies.G-SESAME10developedanewalgorithmtomeasurethefunctionalsimilarity.Thewebtoolitofferedonlycanbeusedtomeasurethefunctionalsimilarityoftwogeneproducts.FunSimMat11calculatedthesimilarityofproteinsinUniProtKB12.Awebsearchenginewasdevelopedtoretrievethefunctionalsimilarityofproteins.Itwouldbehelpfulifatoolcouldassistbiologiststocomparethefunctionalrelationshipofinterestedgeneswithwholegenomegeneproducts.However,genome-widerelationshipanalysiscouldnotbecarriedoutinordinarycomputingservers.Itwouldcostdozensofhourseveninhighperformancecluster.Wedevelopedanonlinesystemforfunctionalrelationshipanalysisofgenome-widegeneproducts.Anall-against-allfunctionalsimilaritycomparisonforgenome-widegeneproductsinhuman,mouseandratwerecomputedpreliminarilybasedonouralgorithms.Threedatabaseswerebuilttointegratethesimilarityscoresrespectively.Basedontheprecomputedsimilarityscores,awebsearchenginewasdevelopedtoretrievethesimilarityscoresdireclty.Someotherrelatedtoolsweredevelopedtoextendtheonlinewebservices.Biologistscanusethesystemeasilytoanalyzethefunctionalrelationshipofgenome-widegeneproducts.II.CONSTRUCTIONANDCONTENTA.DataSetsTherawdataadoptedtocalculatethesimilarityweredirectlyfromtheannotationpackagesofR/Bioconductorproject13,14.Forexample,thepackagesorg.Hs.eg.db,org.Mm.eg.dbandorg.Rn.eg.dbcontainedtheGOannotationdataofgeneproductsinhuman,mouseandratrespectively.ThepackagesweredescribedinthetableI.AlltheseGOrelatedpackageswerebuiltbyBioconductorprojectaccordingtothelatestversionofGOdatabasein2009March.TheannotationdataofprobeIDsofdifferentmicroarrayplatformswerealsofromtheannotationpackagesinBioconductor.B.Implement1)Algorithm:Threedatabasesintegratedallsimilarityscoresofgenome-widegeneproductsinhuman,mouseandratrespectively.Weproposedanovelalgorithmtomeasuretherelationship.Statisticalmodelwasbuiltaccordingtothecommoninformationoftheannotationtermsbetweentwogeneproducts.TheGOprovidedthreestructuredvocabularies(ontologies)todescribegeneproductsintermsoftheirassociatedbiologicalprocesses(BP),cellularcomponents978-1-4244-4713-8/10/$25.002010IEEEFig.1.Functionalsimilaritysearchforgeneproducts.TABLEIDATASETSADOPTEDINTHEDATABASESAnnotationpackagesSpiecesRawdataorg.Hs.eg.dbHumanGOannotation；M.Mm.eg.dbMousedittoorg.Rn.eg.dbRatdittoorg.Hs.sp.dbHumanProteinidentifierstoEntrezIDsorg.Mm.sp.dbMousedittoorg.Rn.sp.dbRatdittoGO.db-GOtermsrelationshipandannotationKEGG.db-AnnotationmapsforKEGGdatabase(CC)andmolecularfunctions(MF).TheGOtermscouldbeconnectedwithchild-parentrelationshipbetweeneachother.ThethreeontologieswerestructuredasDirectedAcyclicGraph(DAG).GOtermswereindifferentlevelsoftheDAG.ThetermslocatedclosetotheleavesofDAGdescribedmorespecificmeanings.Thesetermscontainedmoreinformationthanthetermslocatedclosetotheroot.Wedefinedaparameter,LevelCoefficient(LC),todenotetheweightoftheinformationofaGOterm.TheLCvaluesofleavesweredefinedas1.Fromchildrentoparents,theLCvaluesgraduallydecreasedastheratiooftheirlevelsintheDAG.Ageneusuallywasannotatedbymorethanoneterminthreeontologies.Theinformationofatermshouldalsocontaintheinformationofitsancestorterms.Thus,thecommontermsbetweentwogeneproductscouldbesummarizedtoacontingencytable.TheLCvaluesasinformationweightsoftermscouldbecountedtothecontingencytable.Therefore,therelationshipoftwogeneproductscouldbemeasuredbystatisticallytestingtheagreementofthecontingencytable.WeadoptedKappavaluetotesttheagreement.Furthermore,theZtestwasusedtotestthesignificantofKappavalue.Whentwogeneproductswerefunctionallyrelated,theKappavaluewouldbecloseto1.2)SimilarityScoresComputation:Therearemorethantenthousandsgeneproductsindifferentspecies.All-against-allcomparisonofallgeneproductsrequiredsolargeamountofcomputingpowerthatordinarycomputerscouldnotfinishthecalculation.Thecomputationaltaskwasseparatedintosmalltasksbydividingtheinputdata.Iftheamountofgenome-widegeneproductsisn,theithcalculationtaskwastocalculatethesimilarityscoresbetweentheithgeneproductandtheonesfromthefirsttotheithgeneproducts.DifferentcalculationtaskswereassignedtodifferentCPUsinahighperformancecluster.Thenthecomputationalresultsweresummarizedtoamatrixofsimilarityscores.ParallelprogramsbasedonRlanguageweredevelopedtorealizethecomputation.RpackagesRmpi15andsnow16providedparallelinterfacestoMPIlibraryoftheclusterenvironment.C.DatabasesThreedatabaseswerecreatedtointegratetheprecomputedsimilarityscoresmatricesofallgeneproductsinhuman,mouseandrat.ThescoresincludedKappavaluesandZscoresbetweeneverytwogeneproducts.Forexample,therewere17482humangeneproducts,thenthescorematrixwiththedimensionof1748217482wouldbestoredinthedatabases.Rlanguage13wereusedtodevelopprogramstoperformthecomputation.Theresultsmatricesweresohugethatitwasdifficulttobestoredinregularrelationaldatabase.Fig.2.Onlinetoolsforfunctionalrelationshipanalysis.Weformattedthelargescorematricesintohundredsofmatriceswithsmallerdimensions.ThenoursystemstoredthematricesdatadirectlyinRbinaryfiles(Rdata).Thevolumeofdatabasefileswasapproximate4gigabytesinsize.ThefiledatabasecouldbeimportedbyRscripts.D.WebsystemThesystemcouldbevisitedthoughinternettoretrieveandanalyzethefunctionalrelationshipofgeneproducts.TheApachehttpserverwasusedtoparsetheHTMLwebpages.Throughthewebserver,theuserscouldsubmittheirdatatothesystemandtheresultswouldbereturnedonthewebpages.Renvironmentwasthebaseofthesystem,whichwasinchargeofdataanalysisandinteractingwiththedatabases.Rapache17asafunctionalmoduleofApache,connectedthewebserverandRenvironment.ThedataandvariablessubmittedbytheuserscouldbetransferredtoRenvironmentviaApache.TheresultsfromRprogramsalsocouldbereturnedtotheusersthroughthewebserver.III.UTILITYANDDISCUSSIONA.WebInterfacesWebinterfacestothedatabaseandanalysistoolsweredeveloped.Asshowninfigure1,ourwebtoolsweredesignedintheconciseanduser-friendlyway.Thesystemprovidedthetoolsoffunctionalsimilaritysearchandclassificationforgeneproducts.Someothertools,suchasgeneenrichmentanalysis,identifierconversionandGOannotation,wereextendedtothesystemtoassistthedataanalysis.DocumentswerealsowrittenintheFAQpagetodescribethetoolsandgiveexamples.B.FunctionalsimilaritysearchforasinglegeneproductThegFSimtoolprovidesafunctiontosearchthemostrelatedgeneproductsforasinglegeneproductinthegenome(Figure1A).SeveralidentifiersofgeneproductsincludingEntrezID,Symbol,UnigeneandSwissProtIDweresupported.Geneproductsinthreespeciesincludinghuman,mouseandratcouldbesearchedinthetool.Thenumberofgeneproductsintheresultscouldbespecified.Thetop100functionallysimilargeneproductswouldbereturnedintheresultsbydefaults.EntrezID,annotatedGOtermsandZscoreswouldbeshowninthesearchresults(Figure1B).GeneproductsannotatedwiththesameGOtermswouldbeputinthesamerow.ThesearchresultscouldalsobedownloadedintheCSV(commaseparatedvalues)formatfile.C.FunctionalsimilarityanalysisforagroupofgeneproductsThegsFSimtoolcouldbeusedtoretrieveandanalyzethefunctionalrelationshipofagroupofgeneproducts(Figure1C).MultipleidentifiersandspeciesofgeneproductsweresupportedinthetoolassameasgFSim.Agroupofformattedgeneproductscouldbesubmittedwiththeseparatorssuchascommas,semicolons,spacesandlinebreaks.AsimilarityscorematrixoftheinputgeneproductswithKappavalueswasshownintheresults.Thesimilarityscorematrixwasalsographicallyvisualized.Aheatmap(Figure1D)demonstratedtheannotatedGOtermsofgeneproducts.ThebluecolorinthegraphdenotedthetheGOtermswereusedtoannotatethecorrespondinggeneproducts.Blackmeantthesetermsdidnotannotatethegeneproducts.Adendrogram(Figure1E)intheresultsshowedthehierarchicalclusteringresultsaccordingtothesimilarityscorematrix.Geneproductswereclassifiedintodifferentgroupsbasedontheirfunctionalrelationship.D.EnrichmentAnalysisGeneenrichmentanalysis18isausefulmethodtodiscoverthespecificfunctionalannotationintheselectedgenesfromthetotal(universe)genes.Asshowninfigure2A,theannotationdatabaseshouldbeselectedfirstly.BP,MFandCContologyofGOdatabaseandKEGGpathwaydatabase19weresupportedinthetool.Thenthep-valueofsignificanttestintheenrichmentanalysisalgorithmcouldbespecified.Thep-valuewas0.05bydefault.Iftheannotationtermwasmorespecificandimportantintheselectedgeneproducts,thetermwouldgetasmallerp-value.Thisvaluecouldbeusedtorestrictthenumberofresults.Iftherewasnoresultintheenrichmentanalysis,abiggerp-valuecouldbeassigned.Agroupofinterestedgeneproductscouldbesubmittedtotheselectedgenes.Theoverallgeneproductsshouldbesubmittedastheuniversegenes.Theanalysisresultsincludethesignificantlyenrichedfunctions,P-values,oddsratio,andannotatedcounts(Figure2B).TheresultscouldalsobedownloadedintheCSVformatfile.Theenrichmentanalysistoolcouldbeusedtoanalysistheresultsoffunctionalsearchforagroupofgeneproducts(gsFSim).E.MicroarrayProbeIDConversionThemicroarrayprobeIDconversiontoolcouldtransfertheprobeIDsfromdifferentmicroaryplatformstoEntrezIDs(Figure2C).Mostofcommercialgenechips,suchasAffymetrix,Agilent,GE(GeneralElectric)andIlluminaweresupported.MicroarrayprobeIDscouldbeconvertedtoEntrezID,thentheIDscouldbesubmittedtotheothertoolstoanalyzethefunctionalrelationship.Therefore,thetoolextendsthesupportedidentifierstypesofgeneproductsinthesystem.F.GOAnnotationAsetofGOtermscouldbesubmittedtotheannotationtooltosearchthedetailedinformationinbatch.AfteragroupofGOtermsweresubmitted,theresultswouldbereturnedincludingthetermnames,definitions,synonymsandLCvaluesindescendingorderofLCvalues.LCdenotedtheweightedinformationofaGOterm.Thusthetermswithmorespecificbiologicalmeaningswouldbeshowninthefrontoftheresults.IV.CONCLUSIONForthepurposeofdevelopingapowerfulanduser-friendlytooltoanalyzethefunctionalrelationshipofgenome-widegeneproducts,wecomputedthefunctionalsimilarityscoresofallgeneproductsinhuman,mouseandratbasedonouralgorithminadvance.Anonlinesystemwasdevelopedonthebaseoftheprecomputedsimilarityscores.Thesystemprovidedagroupoftoolstoretrievethefunctionalsimilarityandanalyzetherelationshipforgenome-widegeneproducts.Ourwebservicesarefreelyavailableat/fsim/index.html.ACKNOWLEDGMENTThisworkwaspartiallysupportedbyChinaMedicalBoardofNewYork,Inc.#03-787.ThecomputingtasksofsimilarityscorematriceswereperformedintheHighPerformanceComputingCenter,PekingUnionMedicalCollege.REFERENCES1L.J.Lu,Y.Xia,A.Paccanaro,H.Yu,andM.Gerstein,“Assessingthelimitsofgenomicdataintegrationforpredictingproteinnetworks.”GenomeRes,vol.15,no.7,pp.945953,Jul2005.2A.Schlicker,C.Huthmacher,F.Ramrez,T.Lengauer,andM.Albrecht,“Functionalevaluationofdomain-domaininteractionsandhumanpro-teininteractionnetworks.”Bioinformatics,vol.23,no.7,pp.859865,Apr2007.3M.E.Futschik,G.Chaurasia,andH.Herzel,“Comparisonofhumanprotein-proteininteractionmaps.”Bioinformatics,vol.23,no.5,pp.605611,Mar2007.4E.A.Adie,R.R.Adams,K.L.Evans,D.J.Porteous,andB.S.Pickard,“Suspects:enablingfastandeffectiveprioritizationofpositionalcandidates.”Bioinformatics,vol.22,no.6,pp.773774,Mar2006.5Y.QuandS.Xu,“Supervisedclusteranalysisformicroarraydatabasedonmultivariategaussianmixture.”Bioinformatics,vol.20,no.12,pp.19051913,Aug2004.6M.Ashburner,C.A.Ball,J.A.Blake,D.Botstein,H.Butler,J.M.Cherry,A.P.Davis,K.Dolinski,S.S.Dwight,J.T.Eppig,M.A.Harris,D.P.Hill,L.Issel-Tarver,A.Kasarskis,S.Lewis,J.C.Matese,J.E.Richardson,M.Ringwald,G.M.Rubin,andG.Sherlock,“Geneontology:toolfortheunificationofbiology.thegeneontologyconsortium.”NatGenet,vol.25,no.1,pp.2529,May2000.7S.Carbon,A.Ireland,C.J.Mungall,S.Shu,B.Marshall,S.Lewis,A.O.Hub,andW.P.W.Group,“Amigo:onlineaccesstoontologyandannotationdata.”Bioinformatics,vol.25,no.2,pp.288289,Jan2009.8D.Binns,E.Dimmer,R.Huntley,D.Barrell,C.ODonovan,andR.Apweiler,“Quickgo:aweb-basedtoolforgeneontologysearching.”Bioinformatics,vol.25,no.22,pp.30453046,Nov2009.9A.Schlicker,J.Rahnenfhrer,M.Albrecht,T.Lengauer,andF.S.Domingues,“Gotax:investigatingbiologicalprocessesandbiochemicalactivitiesalongthetaxonomictree.”GenomeBiol,vol.8,no.3,p.R33,2007.10Z.Du,L.Li,C.-F.Chen,P.S.Yu,andJ.Z.Wang,“G-sesame:webtoolsforgo-term-ba