The Research On Functional Module Detection Algorithms Of Co-regulatory Network

In recent years,with the completion of the human genome rough draft and genetic sequencing project,a large number of genomics data has been generated,including gene expression data,gene sequence data,and molecular regulatory relationship and so on.These data could help researchers understand lifecycle,cellular process,gene mutations,and the pathogenesis of complex disease.Meanwhile,with the development of computer science,studying the regulation,driving pathway and cancer process in vivo at network level has become a hot research issue.And as a basic unit achieving cell function,the functional modules assist to understand the biological regulatory mechanism.Based on the cancer microRNA,transcription factor and gene co-regulatory network,two functional module detection algorithms are proposed in this thesis:Firstly,considering that the current biological functional module mining algorithms rarely contain transcription factor,ignoring the synergistic regulatory of the two(miRNA and transcription factor)regulators on target genes.A functional module detection method based on density-based peaks clustering is proposed.Firstly,based on the Pearson correlation network,robust principal component analysis is used to find the sparse matrix with high correlation edges and the low rank matrix including the original network structure.Then,the topological similarity and density of each points are calculated with the sparse matrix,the density peaks are found as the module clustering centers.In order to avoid the deviation of manually selection cluster centers in the decision graph,the turning point with threshold is used to choose centers.Finally,considering the particularity of biological network,the topological similarity and the biological correlation strength are balanced to calculate the assigned probability,afterwards the remaining nodes will be assigned to the module with higher probability.The experimental results show that compared with SNMNMF,DP and OSC,CRMDP can identify more significant functional modules ranging from topological characteristics to biological enrichment.Secondly,considering that only taking expression profiles as input data can not effectively contain more valid biological information,and CRMDP requires to select overlapping parameter.A neighborhood inflation co-regulatory module identification algorithm APNICRM is proposed,which can be regarded as a three-stage overlappingcommunity discovery problem.The algorithm combines expression profile data and prior regulatory relationships,and constructs a miRNA/TF/mRNA co-regulatory network based on regularized least squares method.Then,the affinity propagation algorithm is adopted to generate original module seeds,and neighborhood similarity merging strategy is used to combine the seeds with higher similarity as candidate seed sets.Finally,a new neighborhood inflation strategy is used to diffuse nodes to form functional modules with optimal fitness function.Compared with the SNMNMF,CRMDP and AP algorithm,the modules identified by APNICRM are more dense and relevant,and the enrichment scores are also higher,which have strong biological significance.