An applicationofkernelmethodstogeneclustertemporalmeta-analysis

The applicationofvariousclusteringtechniquesforlarge-scalegene-expressionmeasurementexperi- ments isawell-establishedmethodinbioinformatics.Clusteringisalsousuallyaccompaniedbyfunctional characterization ofgenesetsbyassessingstatisticalenrichmentsofstructuredvocabularies,suchasthe gene ontology(GO)[GeneOntologyConsortium.Thegeneontology(GO)projectin2006.NucleicAcids Research (Databaseissue),vol.34;2006.p.D322–6].Ifdifferentclustersaregeneratedforcorrelated experiments, amachinelearningsteptermed cluster meta-analysis may beperformed,inordertodiscover relations amongthecomponentsofsuchsets.Severalapproacheshavebeenproposed:inparticular, kernel methods may beusedtoexploitthegraphicalstructureoftypicalontologiessuchasGO.Following up theformulationofsuchapproach[MericoD,ZoppisI,AntoniottiM,MauriG.Evaluatinggraphkernel methods forrelationdiscoveryinGO-annotatedclusters.In:KES-2007/WIRN-2007,PartIV,Lecturenotes in artificialintelligence,vol.4694.Berlin:Springer;2007.p.892–900;ZoppisI,MericoD,AntoniottiM, Mishra B,MauriG.DiscoveringrelationsamongGO-annotatedclustersbygraphkernelmethods.In:Pro- ceedings ofthe2007internationalsymposiumonbioinformaticsresearchandapplications.Lecturenotes in computerscience,vol.4463.Berlin:Springer;2007],inthispaperwediscuss,fromaninformation- theoretic pointofview,furtherresultsaboutitsapplicabilityanditsperformance.