模块网络：变量分组与条件概率分布共享机制.pptxVIP

ModuleNetworks

BMI/CS576MarkCravencraven@biostat.wisc.eduDecember2007

ModuleNetworksMotivationsetsofvariablesoftenhavethesamebehaviorconsiderthissimplestockexamplewecangroupvariablesintomodules,havethemembersofamodulesharethesameCPDFigurefromSegaletal.,UAI,2003.

ModuleNetworksamodulenetworkisdefinedbyaspecifiednumberofmodulesanassignmentofeachvariabletoamoduleasharedCPDforthevariablesineachmodulethelearningtaskthusentails*determiningtheassignmentofvariablestomodulesinducingaCPDforeachmodule*assumingwe’regiventhenumberofmodules

ModuleNetworks:IdentifyingRegulatoryModulesandtheirCondition-SpecificRegulatorsfromGeneExpressionData.E.Segaletal.,NatureGenetics34(2):166-176,2003given:geneexpressiondatasettheirmethodidentifies:setsofgenesthatareco-expressed(assignmenttomodules)a“program”thatexplainsexpressionprofileforeachsetofgenes(CPDforeachmodule)

ARegulationProgramsupposewehaveasetof(8)genesthatallhaveintheirupstreamregionsthesameactivator/repressorbindingsites

TheRespirationandCarbonModule

RegulationProgramsasCPDseachoftheseregulationprogramsisactuallyaCPDrepresentedusingatreeinternalnodesaretestsoncontinuousvariablesleavescontainconditionaldistributionsforthegenesinthemodule,representedbyGaussiansisHAP40.1P(expression|leaf)

TreeCPDsinGeneModuleNetworkstheparentsofeachmodulecanbeothermodulesinthisstudy,Segaletal.limitparentstoasetofcandidateregulatorgenes(genesknowntobetranscriptionfactorsandsignalingcomponents)

ModuleNetworkLearningProceduregiven:expressionprofileforeachgene,setofcandidateregulatorgenesinitializemoduleassignmentsbyclusteringexpressionprofilesrepeatuntilconvergence structuresearchstep: foreachmodule learnaCPDtreeusingsplitsoncandidateregulators moduleassignmentstep: repea