The Research On Functional Module Identification Algorithms Of Co-regulatory Networks Based On Multi-source Data

With the development and application of next-generation high-throughput sequencing technology,omics data such as genomics,transcriptomics data has been rapid growth.On the one hand,the large volume of omics data provides new opportunities for studying the functions of biomolecules from network level.On the other hand,it still remains a big challenge to utilize and integrate various kinds of omics data and identify valuable biological information.The co-regulatory network which is composed by miRNA,transcription factor and gene is one of the current hot research topics in systematic bioinformatics.Functional modules,the main carrier of cell functions in co-regulatory network,is significant for understanding the molecular mechanism of the organism and complicated diseases etiopathogenesis.In the thesis,considering the property of co-regulatory network,we proposed two methods to identify functional modules in co-regulatory network by integrating multiple types of genomic data.The current regulation network less regard transcription factor as regulator which regulating target genes.So existing methods may not identify transcription factors and its regulations in co-regulatory network effectively.Here we propose a novel algorithm based on non-negative matrix factorization to discover functional modules called SNCoNMF.The algorithm first adopts joint non-negative matrix factorization to integrate expression profiles of miRNA,transcription factor and gene;furthermore,the co-regulatory network-regularized constraints of miRNA-gene,TF-gene and gene-gene are incorporated with joint non-negative matrix factorization,to ensure that factors with edges share higher probability of belonging to same module;we finally add sparse penalty term to object function in consideration of the sparsity of co-regulatory network and non-negative matrix.The functional modules predicted by SNCoNMF from human co-regulatory network exhibit more transcription factors and explain the cooperative regulation between miRNAs and transcription factors more clearly.Given the fact that the co-regulatory network is sparse and functional modules predicted by SNCoNMF bear small density,we propose a new heuristic algorithm named NPWCN based on the network node permanence.We first construct more reliable weighted network based on LASSO with miRNA,transcription factor and gene expression profiles;then identify key regulators in weighted co-regulatory network by linear programming and take them as seed nodes in consideration of key regulators taking dominant position in co-regulatory network;we finally predict functional modules by expanding the neighbor nodes of seed nodes based on greedy policy where taking average node permanence as object function.The experimental results indicate that the identified modules by NPWCN achieve higher density and biologically enriched more closely.