Construction And Analysis Of MicroRNA-mediated Regulatory Networks In Cancer

MicroRNAs(miRNA)are a class of ~22nt endogenous non-coding RNAs that bind to 3’ UTR of the target genes and inhibit their expression post-transcriptionally.A large number of studies are devoted to exploring the functional mechanisms of miRNA in the development and progression of diseases,with particular emphasis on their role in cancers.In general,miRNAs perform their regulatory effects by binding to target genes,but usually other regulatory factors(such as transcription factor,TF and long noncoding RNA,lncRNA)are involved in coordinated regulation or blocking their regulation on their targets.That complicates the regulatory mechanism.Studies have shown that miRNAs are potentially involved in multiple regulatory relationships,such as miRNA binding to target genes,thereby constructing a miRNA-mRNA regulatory network;TF and miRNA binding to the same target genes,thereby constructing a TFmiRNA-mRNA co-regulatory network;lncRNA and mRNA competitively binding to the same miRNAs,thereby constructing a lncRNA-miRNA-mRNA competitive triple regulatory network.More importantly,many studies have found that miRNA may involve almost all biological processes and pathways,including carcinogenesis.Due to the complexity of regulatory mechanisms coupled with the specificity of different cancers,the construction and analysis of miRNA-mediated regulatory networks has been a challenging subject.Therefore,in this thesis,we elucidate the functional significance of miRNA by deducing its context-specific regulatory role.According to the different research background and purpose,we propose different computational methods to construct cancer-specific regulatory networks and to discover some important sub-networks,and further to analyze the corresponding topological features and functional characteristics.The main research work of this thesis is as follows.(1)Considering that co-expressed genes are more likely to have similar regulatory mechanisms and functions,we designed a coexpression correlation-based miRNA regulatory module identification algorithm named CoModule.The algorithm first uses the miRNA expression profile data to construct a miRNA co-expression similarity network,and then employs a computational method to obtain miRNA clusters with high co-expression correlation.The CoModule and the comparison algorithms were applied to the ovarian cancer dataset.The results showed that compared with the comparison algorithms,the modules identified by CoModule have tighter regulatory connections and exhibit stronger miRNA-mRNA negative correlation expression coefficient.In addition,miRNAs in the same module are significantly enriched in the corresponding miRNA family,while miRNA target genes exhibiting consistent functional enrichment.In general,the miRNA regulatory modules identified by CoModule have good performance in both topology and biological significance.(2)Since the existing module identification algorithm cannot identify the miRNA functional modules with prognostic significance,we proposed a priori clinical information-based miRNA prognostic module identification method called ProModule.The algorithm first uses the priori clinical information of tumor samples to select the miRNA individuals associated with the prognosis,and then employs a clustering method to systematically identify miRNA modules with statistical prognostic significance.We applied ProModule and the comparison method to three data sets for testing.The results indicated that the miRNA module exhibits a stronger prognostic value than the individual.Compared with the comparison method,the miRNA module identified by ProModule exhibits a more consistent miRNA-miRNA co-expression correlation coefficient,and the miRNA target genes in the module are significantly enriched in some cancer-related biological functional processes and pathways.(3)In order to explore the synergistic regulation mechanism of TF and miRNA targeting common genes,we developed a bipartite graph-based TF-miRNA coregulatory subnetwork identification method called BiModule.This method first reconstructs a cancer-specific TF-miRNA co-regulatory network by integrating the gene expression profiles and binding site information,and then employs a biclique modularity strategy to mine the TF-miRNA co-regulatory subnetworks.We applied BiModule and the comparison method to the cervical cancer dataset.The results showed that,compared with the comparison method,the TF-miRNA co-regulatory sub-network identified by BiModule exhibits a denser connection and a stronger expression correlation.At the same time,a portion of sub-networks show statistically significant prognostic relevance,and their target genes also exhibit significant biofunctional enrichment.(4)Since the current computational identification of lncRNA associated competing triplet is not accurate enough,we proposed a statistical-based lncRNA-miRNA-mRNA competing triplet identification method named lncTriplet.The algorithm first designs a new computational scheme to simulate the ceRNA regulation process,and then uses the null hypothesis test to discover the lncRNA-miRNA-mRNA competing triplets.The lncTriplet was applied to breast cancer data set.The results showed that the lncRNAmiRNA-mRNA competing triplets identified by lncTriplet are consistent with the ceRNA regulation mechanism,and its corresponding lncRNA and mRNA have a more consistent co-expression similarity and differential expression pattern.