Evolutionary SearchofBiclusters byMinimalIntrafluctuation
Rautl Giraldez, Federico Divina, Beatriz Brígido Pontes · 2007
Biclustering techniques aimatextracting signif- icantsubsets ofgenesandconditions frommicroarray gene expression data. Thiskindofalgorithms ismainly basedontwo keyaspects: thewayinwhichtheydealwithgenesimilarity across theexperimental conditions, thatdetermines thequality ofbiclusters; andtheheuristic orsearchstrategy usedfor exploring thesearch space. A measurethatisoften adopted forestablishing thequality ofbiclusters isthemeansquared residue. Thismeasurehasbeensuccessfully usedinmany approaches. However, ithasbeenrecently proventhatthe meansquared residue fails torecognize somekindofbiclusters asquality biclusters, mainly duetothedifficulty ofdetecting scaling patterns indata.Inthiswork,we proposea novel measurefortrying toovercome thisdrawback. Thismeasure isbasedon theareabetweentwocurves. Suchcurvesare built fromthemaximumandminimumstandardized expression valuesexhibited foreachexperimental condition. Inorder totesttheproposed measure, we haveincorporated itinto a multiobjective evolutionary algorithm. Experimental results confirm theeffectiveness ofourapproach. Thecombination of themeasurewepropose withthemeansquared residue yields results thatwouldnothavebeenobtained ifonlythemean squared residue hadbeenused.