研究每個人的單一核甘酸多型性(SNP)的差異課件

ApplicationofSupportVectorMachinetodetectanassociationbetweenadiseaseortraitandmultipleSNPvariationsAuthor:GeneKim,MyungHoKimAdvisor:Dr.HsuGraduate:Ching-WenHongApplicationofSupportVector1Outline1.Motivation2.Objective3.What’sSNP(singlenucleotidepolymorphism)4.HowtofindSNPvariations5.AreviewofSupportVectorMachine6.ArepresentationofmultipleSNPvariationsasavector7.Themarks

8.InseparableCase9.Testresultswithclinicaldata10.PersonalopinionOutline1.Motivation2Motivation研究每個人的「單一核甘酸多型性」(SNP)的差異,可以幫助了解致病基因,甚至預(yù)測藥物對個人是否具有療效,進一步設(shè)計量身訂做藥物,對新藥的開發(fā)有極大的影響。SNP的研究是後基因時代生技產(chǎn)業(yè)發(fā)展的主要趨勢。Motivation研究每個人的「單一核甘酸多型性」(SNP3ObjectiveWecanpresentamethodofdetectingwhetherthereisanassociationbetweenmultipleSNPvariationsandatraitordisease.ThemethodexploitstheSupportVectorMachine(SVM)whichhasbeenattractinglotsofattentionsrecently.ObjectiveWecanpresentameth4What’sSNP何謂SNP(單一核甘酸多型性)雖然同種生物其染色體差異極小,但平均1000個鹼基對(basepair)就有一個發(fā)生突變,這些變異稱為SNP,是造成每個人對藥物的敏感性不同、血型不同、身高等等的原因。此外,SNP也和癌癥、心血管疾病、自體免疫等等疾病有關(guān)。目前國內(nèi)賽亞基因和臺大醫(yī)院合作,正從事C型肝炎SNP研究,試圖找出病患的SNP,以預(yù)測藥物是否對病人有效。What’sSNP何謂SNP(單一核甘酸多型性)5What’sSNP

AgeneticmarkerisM1,M2,…intheDNAThedifferentvariantsofDNAthatdifferentpeoplehaveatthemarkerarealleles,denotedby1,2,3..,Thenumberofallelespermarkerissmall:typicallylessthanten(forcalledmicrosatellitemarker)orexactlytwo(forcalledSNPs).What’sSNPAgeneticmarkeris6HowtofindSNPvariationsTheproblemofdeterminingwhetherasetofSNPvariationcauseaspecificdiseaseortraitcouldbeformulatedasfollows.Foragivendiseaseortrait,1.ForeachsetofSNPvariations,finditsrepresentationasavectorinaEuclideanspace.(haplotypedata,clinicaldata,….wewilldiscussthisinthepage9)2.GetasystematicwayofdistinguishingSNPgenotypeofnormalpeoplefromonesofpeoplewiththediseaseortrait.WewillusetheSupportVectorMachine(SVM)toseparateSNPvectorsintotwogroups(normal,sick).HowtofindSNPvariationsThe7AreviewofSupportVectorMachineWhatisaSVM?afamilyoflearningalgorithmforclassificationofobjectsintotwoclasses.Input:atrainingset{(x1,y1),…,(xl,yl)}ofobjectxiE?(n-dimvectorspace)andtheirknownclassesyiE{-1,+1}.Output:aclassifierf:?→{-1,+1}.whichpredictstheclassf(x)forany(new)objectxE?AreviewofSupportVectorM8AreviewofSupportVectorMachine(1).LinearSVMforseparabletrainingsets:atrainingsetS={(x1,y1),…,(xl,yl)},xiE?,yiE{-1,+1}.AreviewofSupportVectorM9AreviewofSupportVectorMachineTheoptimalhyperplaneisdefinedbythepair(w,b).SolvethelinearprogramproblemMin?║w║2

st.yi(xi·w+b)-1≥0,i=1,…,lThisisaclassquadratic(convex)programAreviewofSupportVectorM10AreviewofSupportVectorMachine(2).LinearSVMfornon-separabletrainingsetsSolvethelinearprogramproblemMin?║w║2+C(∑εi),cisaextremelargevalueS.t.yi(xi·w+b)-1+εi

≥0,εi

≥0,0≤αi≤c,i=1,…,lAreviewofSupportVectorM11ArepresentationofmultipleSNPvariationsasavector

SchemeGiveneachdiseaseortrait,andacollectionofSNPdatawhichdependingongenotypeinaconsistentway.(haplotype,clinicaldata):7step1.Assumethatthereisnoenvironmentalfactor.2.SNPlocationsareassumedtobeknowforthediseaseortrait.3.AssumethereisareferenceSNPdata.(goodhealthrecords)4.Bygivingscoresbasedondifferencefromthereferencedata,assignavectortoeachSNPdata.ArepresentationofmultipleS12ArepresentationofmultipleSNPvariationsasavectorThedimensionofvectoristhenumberofSNPstotherelateddiseaseortrait.5.Atrainingsetischosenforthediseaseortrait,inotherwords,SNPgenotypedataofnormalandsickpopulation.6.ByusingStep4,computetheSNPvectorsofthetrainingdataset﹛(xi,yi)﹜,xiisaSNPdata,yi=1(sick)or-1(normal),7.UsetheSVMtogetahyperplanedividingintotwogroups(sick,normal)ArepresentationofmultipleS13Theremarks1.ThereferencedatacanbebuiltbycollectingSNPgenotypesfromthehealthynormalpopulation.2.Thehyperplaneobatinedcanbeconsideredasacriterion,and,givenanewdataset,itcanbeusedfortestingwhetherthepersonofthedataissusceptibletothediseaseortrait.3.RepresentationofanobjectasavectormightbecriticalformakingusetheSVM.Howtomakedomainknowledgecontainedinvectorrepresentationsisoneofthemajorissues.4.Theideaofdifferencescoringcouldbeappliedtootherdatasets(visualdatasuchasX-rayorMRIimage,…),inparticular,tohaplotypedataandtofindoutalinkageamongSNPtothediseaseortrait.5.OnceagroupofSNPpatternsareidentified,itcancomputecontributionscoreofeachofthoseSNPtothediseaseortrait.Theremarks1.Thereferenceda14InseparableCaseFortheinseparablecase,theiterateduseofSVMenablesustodivideacollectionoflabelledofvectorsintoseveralclusteringgroups.1.Setathresholdvalue.Say,80%.2.UseSVMtoseparateacollectionoflabelledofvectorsintotwogroupsA,B.3.Checkifthegroupscontainmorethan80%ofeither1or-1labeledvectors.SupposeAisnotsuchone.ThenuseSVMtoAagaintotwosubgroups.4.Repeatthisprocedureuntileachsubgrouphasamajorityofmorethan80%.5.Foreachsubgroup,figureoutarange.

InseparableCaseFortheinsepa15TestresultswithclinicaldataTheclinicaldataisacardio-patientrecordsdataset:Height,age,sex,weight,etnicbackground,medicalhistory,birthplace,bloodpressure(systolicanddiastolic),Liqidmeasurementsetcarenumericalizedand+1:apatientwithheartattack,strokeorheartfailure,otherwise-1WeusedThorstenJoachims’implementationofSVM.Testresultswithclinicaldat16PersonalopinionApplicationofSVMiseffective,Butitisdifficulttosolvenonlinearproblem.Howtomakedomainknowledgecontainedinvectorrepresentationsisoneofthemajorissues.

17PersonalopinionApplicationofApplicationofSupportVectorMachinetodetectanassociationbetweenadiseaseortraitandmultipleSNPvariationsAuthor:GeneKim,MyungHoKimAdvisor:Dr.HsuGraduate:Ching-WenHongApplicationofSupportVector18Outline1.Motivation2.Objective3.What’sSNP(singlenucleotidepolymorphism)4.HowtofindSNPvariations5.AreviewofSupportVectorMachine6.ArepresentationofmultipleSNPvariationsasavector7.Themarks

8.InseparableCase9.Testresultswithclinicaldata10.PersonalopinionOutline1.Motivation19Motivation研究每個人的「單一核甘酸多型性」(SNP)的差異,可以幫助了解致病基因,甚至預(yù)測藥物對個人是否具有療效,進一步設(shè)計量身訂做藥物,對新藥的開發(fā)有極大的影響。SNP的研究是後基因時代生技產(chǎn)業(yè)發(fā)展的主要趨勢。Motivation研究每個人的「單一核甘酸多型性」(SNP20ObjectiveWecanpresentamethodofdetectingwhetherthereisanassociationbetweenmultipleSNPvariationsandatraitordisease.ThemethodexploitstheSupportVectorMachine(SVM)whichhasbeenattractinglotsofattentionsrecently.ObjectiveWecanpresentameth21What’sSNP何謂SNP(單一核甘酸多型性)雖然同種生物其染色體差異極小,但平均1000個鹼基對(basepair)就有一個發(fā)生突變,這些變異稱為SNP,是造成每個人對藥物的敏感性不同、血型不同、身高等等的原因。此外,SNP也和癌癥、心血管疾病、自體免疫等等疾病有關(guān)。目前國內(nèi)賽亞基因和臺大醫(yī)院合作,正從事C型肝炎SNP研究,試圖找出病患的SNP,以預(yù)測藥物是否對病人有效。What’sSNP何謂SNP(單一核甘酸多型性)22What’sSNP

AgeneticmarkerisM1,M2,…intheDNAThedifferentvariantsofDNAthatdifferentpeoplehaveatthemarkerarealleles,denotedby1,2,3..,Thenumberofallelespermarkerissmall:typicallylessthanten(forcalledmicrosatellitemarker)orexactlytwo(forcalledSNPs).What’sSNPAgeneticmarkeris23HowtofindSNPvariationsTheproblemofdeterminingwhetherasetofSNPvariationcauseaspecificdiseaseortraitcouldbeformulatedasfollows.Foragivendiseaseortrait,1.ForeachsetofSNPvariations,finditsrepresentationasavectorinaEuclideanspace.(haplotypedata,clinicaldata,….wewilldiscussthisinthepage9)2.GetasystematicwayofdistinguishingSNPgenotypeofnormalpeoplefromonesofpeoplewiththediseaseortrait.WewillusetheSupportVectorMachine(SVM)toseparateSNPvectorsintotwogroups(normal,sick).HowtofindSNPvariationsThe24AreviewofSupportVectorMachineWhatisaSVM?afamilyoflearningalgorithmforclassificationofobjectsintotwoclasses.Input:atrainingset{(x1,y1),…,(xl,yl)}ofobjectxiE?(n-dimvectorspace)andtheirknownclassesyiE{-1,+1}.Output:aclassifierf:?→{-1,+1}.whichpredictstheclassf(x)forany(new)objectxE?AreviewofSupportVectorM25AreviewofSupportVectorMachine(1).LinearSVMforseparabletrainingsets:atrainingsetS={(x1,y1),…,(xl,yl)},xiE?,yiE{-1,+1}.AreviewofSupportVectorM26AreviewofSupportVectorMachineTheoptimalhyperplaneisdefinedbythepair(w,b).SolvethelinearprogramproblemMin?║w║2

st.yi(xi·w+b)-1≥0,i=1,…,lThisisaclassquadratic(convex)programAreviewofSupportVectorM27AreviewofSupportVectorMachine(2).LinearSVMfornon-separabletrainingsetsSolvethelinearprogramproblemMin?║w║2+C(∑εi),cisaextremelargevalueS.t.yi(xi·w+b)-1+εi

≥0,εi

≥0,0≤αi≤c,i=1,…,lAreviewofSupportVectorM28ArepresentationofmultipleSNPvariationsasavector

SchemeGiveneachdiseaseortrait,andacollectionofSNPdatawhichdependingongenotypeinaconsistentway.(haplotype,clinicaldata):7step1.Assumethatthereisnoenvironmentalfactor.2.SNPlocationsareassumedtobeknowforthediseaseortrait.3.AssumethereisareferenceSNPdata.(goodhealthrecords)4.Bygivingscoresbasedondifferencefromthereferencedata,assignavectortoeachSNPdata.ArepresentationofmultipleS29ArepresentationofmultipleSNPvariationsasavectorThedimensionofvectoristhenumberofSNPstotherelateddiseaseortrait.5.Atrainingsetischosenforthediseaseortrait,inotherwords,SNPgenotypedataofnormalandsickpopulation.6.ByusingStep4,computetheSNPvectorsofthetrainingdataset﹛(xi,yi)﹜,xiisaSNPdata,yi=1(sick)or-1(normal),7.UsetheSVMtogetahyperplanedividingintotwogroups(sick,normal)ArepresentationofmultipleS30Theremarks1.ThereferencedatacanbebuiltbycollectingSNPgenotypesfromthehealthynormalpopulation.2.Thehyperplaneobatinedcanbeconsideredasacriterion,and,givenanewdataset,itcanbeusedfortestingwhetherthepersonofthedataissusceptibletothediseaseortrait.3.RepresentationofanobjectasavectormightbecriticalformakingusetheSVM.Howtomakedomainknowledgecontainedinvectorrepresentationsisoneofthemajorissues.4.Theideaofdifferencescoringcouldbeappliedtootherdatasets(visualdatasuchasX-rayorMRIimage,…),inparticular,tohaplotypedataandtofindoutalinkageamongSNPtothediseaseortrait.5.OnceagroupofSNPpatternsareidentified,itcancomputecontributionscoreofeachofthoseSNPto