Rapidly identifying protein complexes is signifi-cant to elucidate the mechanisms of macromolecular interac-tions and to further investigate the overlapping clinical man-ifestations of diseases.To date,existing computational meth-ods majorly focus on developing unsupervised graph cluster-ing algorithms,sometimes in combination with prior biological insights,to detect protein complexes from protein-protein in-teraction (PPI) networks.However,the outputs of these meth-ods are potentially structural or functional modules within PPI networks.These modules do not necessarily correspond to the actual protein complexes that are formed via spatiotemporal ag-gregation of subunits.In this study,we propose a computational framework that combines supervised learning and dense sub-graphs discovery to predict protein complexes.The proposed framework consists of two steps.The first step reconstructs genome-scale protein co-complex networks via training a su-pervised learning model of 12-regularized logistic regression on experimentally derived co-complexed protein pairs;and the second step infers hierarchical and balanced clusters as com-plexes from the co-complex networks via effective but com-putationally intensive k-clique graph clustering method or ef-ficient maximum modularity clustering (MMC) algorithm.Em-pirical studies of cross validation and independent test show that both steps achieve encouraging performance.The pro-posed framework is fundamentally novel and excels over ex-isting methods in that the complexes inferred from protein co-complex networks are more biologically relevant than those in-ferred from PPI networks,providing a new avenue for identify-ing novel protein complexes.

Suyu MEI

Software College,Shenyang Normal University,Shenyang 110034,China

计算机科学前沿

[1]Suyu MEI-.A framework combines supervised learning and dense subgraphs discovery to predict protein complexes)[J].计算机科学前沿,2022(01):160-173

ifestations,pervised,putationally,excels

framework,combines,learning,dense,subgraphs,discovery,predict,protein,complexes,Rapidly,identifying,signifi,cant,elucidate,mechanisms,macromolecular,interac,further,investigate,overlapping,clinical,diseases,To,existing,computational,majorly,focus,developing,unsupervised,algorithms,sometimes,combination,prior,insights,detect,from,teraction,PPI,networks,However,outputs,these,are,potentially,structural,functional,modules,within,These,do,not,necessarily,correspond,actual,that,formed,via,spatiotemporal,gregation,subunits,In,this,study,proposed,consists,steps,first,reconstructs,genome,scale,training,model,regularized,logistic,regression,experimentally,derived,complexed,pairs,second,infers,hierarchical,balanced,clusters,effective,but,intensive,clique,clustering,ficient,maximum,modularity,MMC,Em,pirical,studies,cross,validation,independent,test,show,both,achieve,encouraging,performance,fundamentally,novel,methods,inferred,more,biologically,relevant,than,those,providing,new,avenue

0.51707