Research On Functional Module Mining Algorithms Of Co-regulatory Networks In Cancers

With the extension and development of high-throughput sequencing technology and DNA microarray technology,a large number of biomolecular interactions omics data have been accumulated.These data dispaly important theoretical and practical value on DNA sequence analysis,gene expression,gene mutations,gene identification,cancer diagnosis and classification,new drug development.At the same time,gene regulatory network composed of these data provides support conditions for the exploration of the biological genetic relationship from the network level.The research of cancer co-regulatory network helps understand the generation process and regulation of oncogenes inside cancer tissues.In this paper,on the basis of analyzing the characteristic of the network topology,two kinds of co-regulatory modules recognition algorithms have been designed based on the miRNA/TF/mRNA expression profiles.Firstly,in view of few studies considered the functional modules which contain TFs and overlapping parts now.A novel computational framework called overlapping spectral clustering(OSC)to systematically detect overlapping CRMs using miRNA/mRNA/TF expression profiles has been proposed.Firstly,empirical Bayes theories are used to improve building more exact relation matrix instead of simple Pearson correlation analysis.In order to ensure the adaptivity of the whole framework,the eigenvalue decomposition(Eigengap)is emploied to determine the module number automatically.Then considering key regulators which may involve in more function modules,a novel overlapping detection approach has been proposed to observe whether the edges between modules(EBMs)can be overlapped by other modules.Comparing with existing methods on breast cancer(BRCA)and ovary cancer(OV)datasets from TCGA,the CRMs identified by OSC are more functionally enriched.Secondly,in view of the increasingly large scale of gene expression data,common module identification algorithms exist many problems,such as large search space,long running time.A novel co-regulatory modules recognition algorithm RMCL-ESA based on improved markov cluster and explosion search algorithm strategy has been proposed.At first,improved markov cluster is adapted to prepoccess gene expression profiles,through three subprocedure: expansion,inflation,prune.This stage can delete regundant genes,making the subsequent processing processmore quickly and saving the storage space.Then,based on the special regulate pattern between regulators(miRNAs/TFs)and target genes,the two-stage explosion search methods has been explored for identifying co-regulatory modules.In the first stage,find the center cluster nodes in the global search space;In the second stage,expand and search from the target genes of center cluster nodes,form co-regulatory function modules,until satisfy local fitness function.Finally,the experimental results show that,CRMs of RMCL-ESA include more significant biological function GO-terms and regulation pathway compared with NJW,SNMNMF algorithms.At the same time,through the study of accumulated experience distribution analysis on GOES and KEGGES,RMCL-ESA can achieve co-regulatory modules with high enrichmrnt degree.And CRMs of RMCL-ESA can separate patients samples and normal samples through survival analysis,which reflect significant biological significance.