The Research On Algorithm Of Identifying Module In The Co-regulatory Network

In the post-gene era,gene regulation is a central topic in the field of systems biology.Exploring and discovering gene regulation mechanisms is of great significance for researchers to understand the life activities and pathogenesis of complex diseases.In multicellular organisms,Transcription factors(TF)and mi RNAs are the two largest families of gene regulators,which are known to work synergistically in the regulation of common target genes at the transcriptional and post-transcriptional levels,respectively.Therefore,it is of wide application value for the diagnosis and treatment of complex diseases to explore important regulatory patterns among TF,mi RNA and m RNA from the perspective of the co-regulatory network composed of these three genes.To this end,based on the TF-mi RNA co-regulatory networks,we propose two co-regulatory modules identification algorithms by combining with expression profile data and various relationship data.Firstly,in the previous modules identification algorithms,most of them model transcription factors as regulatory targets while ignoring the regulat ion of transcription factors,and fail to effectively intergrate multiple types of data.We proposed an ensemble clustering strategy based computational framework named Tri Module to recognize TF-mi RNA co-regulatory modules.We first calculate the expression correlation for each type of gene bases on the maximum information coefficient analysis,and further employ local error discovery rate to extract relatively strong coefficients to derive co-expression correlation matrix.Then,we performe spectral clustering algorithm independently for each matrix,and transform the initial clustering results into the input of the second clustering to construct a novel heterogeneous network that integrates the regula tory interactions and expression profiles.Finally,we perform spectral clustering again on the constructed network to identify TF-mi RNA co-regulatory modules.Compared with comparison algorithms,the results show that the modules identified by Tri Module exhibit a stronger expression correlation and biological function significance.Through survival analysis,numerous modules are found to have significant prognostic relevance.Secondly,in the view of the sparseness problem of constructing the co-regulation network based on the regulatory relationship data and the non-overlapping problem based on the results obtained by the conventional clustering algorithm s,we proposed a link features based algorithm named Link Module to recognize TF-mi RNA co-regulatory modules.We first construct the co-regulatory network by integrating the experimentally verified regulatory relationship and m RNA interaction data,and further extract higher-order structural relationships in the network based on representation learning methods,to reduce the impact of the co-regulatory network sparsity.Then,considering that genes with expression similarity tend to have similar biological functions,we extract expression similarity features based on expression profile data,and integrate the two extracted features to represent gene features.Finally,we construct the link features based on the obtained gene features,and subsequently perform the DBSCAN clustering algorithm on link features to identify overlapping TF-mi RNA co-regulatory modules.The experimental results show that,the co-regulatory modules identified by the Link Module exhibit advantages in terms of topology characteristics and biological functions.In addition,the case study further illustrates the correlation between the mod ules identified by Link Module and the corresponding cancer.