联合聚类非线性相关的时序基因表达数据.pdfVIP

计算 机研 究 与发展 ISSN 1000—1239／CN 11-1777／TP JournalofComputerResearchandDevelopment 45(11)：1865—1873，2008 联合聚类非线性相关的时序基因表达数据 闫雷鸣 孙志挥 吴英杰 张柏礼 (东南大学计算机科学与工程学 院 南京 210096) (yan — leiming：@163．corn) BiclusteringNonlinearlyCorrelatedTimeSeriesGeneExpressionData YanLeiming，SunZhihui，WuYingjie，andZhangBaili (SchoolofComputerScience＆Engineering，SoutheastUniversity，Nanjing210096) Abstract The biclustering algorithms focus on clustering correlated patterns in sub—spaces． However，mostofthebiclusteringalgorithmsnowadaysaddressonlythelinearlycorrelatedpatternor acertain linearly similarpattern，leaving thenonlinearly correlatedpatternsuntouched，which are oftenhiddeninagreatmanyofrealdatasets．Inthispaper，anovelbiclusteringalgorithm calledM I— TSB isproposedto findandreportallnonlinearlycorrelatedpatternsin timeseriesgeneexpression data．Itfirstdeducesan efficientcalculatingformulaofquadraticmutualinformationwith matrix theory，andthen basedon thequadraticmutualinformation and slidingwindow technology，a time seriesdatanonlinearlysimilarmodelandasimplegeneralsuffixtreevariationversionareintroduced． Usingsuffixtreeasindex structure，theM I～TSB algorithm exploresallofbiclusterseffectivelyand efficiently．Comparedwithgeneralbiclusteringalgorithms，theability ofdiscovering thenonlinearly correlatedpatterns in sliding window isone of themost importantadvantagesof the M I—TSB algorithm．-Additionally，experimentsonrealgeneexpressiondatasetandsyntheticdatasetshow that theM I—TSB algorithm successfully discoverssomenonlinearly correlatedpatternswhich can notbe found by other ordinary biclustering algorithms． Besides， gene annotating by gene ontology demonstratesthattheM I—TSBalgorithm canfindbiologicallymeaningfulresults． Keywords quadraticmutualinformation；nonlinearcorrelation；bielustering；bioinformactics；gene expressiondata 摘 要 为聚类非线