Optimization models to identify key RNA regulatory modules in cancer

Micro RNAs (miRNAs) are known as the important components of RNA silencing and post-transcriptional gene regulation, and they interact with messenger RNAs (mRNAs) either by degradation or by translational repression. miRNA alterations have a significant impact on the formation and progression of human cancers. Additionally, finding primary tissue of tumors is a fundamental property of a precise treatment. Therefore, employing computational methods with the ability to identify both tissue-specific and cohort-specific miRNA-mRNA regulatory modules is fast becoming a crucial tool in cancer biology. In this thesis, we first identified regulatory modules of 32 cancer types using a sparse multivariate factor regression (SMFR) model on matched miRNA and mRNA expression profiles of more than 9,000 primary tumors. We used an algorithm that decomposes the coefficient matrix into two low-rank matrices with separate sparsity-inducing penalty terms on each. Although, our solution significantly outperformed another decomposition-based approach in terms of normalized root mean squared error, the interpretability of the algorithm was not satisfying, due to the low sparsity level of solutions. Therefore, we proposed a single task two-step framework to model miRNA-mRNA relationships and identify cancer-specific modules between miRNA and mRNA from their matched expression profiles. We first estimated the regulatory matrix between miRNA and mRNA expression profiles by solving multiple linear programming problems. We then formulated a unified regularized factor regression (RFR) model that simultaneously estimates the effective number of modules (i.e., latent factors) and extracts modules by decomposing regulatory matrix into two low-rank matrices. Our RFR model groups correlated miRNAs together and correlated mRNAs together, and controls sparsity levels of both matrices. These attributes lead to interpretable results with high predictive performance. To find the biological relevance of our approach, we performed functional gene set enrichment and survival analyses. A large portion of the identified modules are significantly enriched in Hallmark, PID and KEGG pathways/gene sets. To validate the identified modules, we also performed literature validation as well as validation using experimentally supported miRTarBase database. By applying proposed single-task algorithm on 32 independent cancer cohorts, interestingly we observed similar patterns in identified miRNAs and mRNAs, also in significance of certain gene sets for specific cohorts. Inspired with the previous results, we attempted to investigate the similarities between mechanism of miRNA-mRNA regulatory modules in different cancer cohorts from the same tissue to improve the predictive performance of the model and extract more detailed and informative solutions. Moreover, detecting commonalities of cohorts from the same tissue, can provide valuable information about the underlying tissue and cohort mechanisms. Hence, we established a multitask learning formulation to identify key tissue- and cohort-specific miRNA-mRNA regulatory modules from their matched expression profiles of tumors. To this end, we proposed a multitask learning sparse regularized factor regression (MSRFR) method to model the sparse relationship between miRNAs and mRNAs, extract tissue- and cohort-specific miRNA-mRNA regulatory modules separately and estimate the number of cohort-specific regulatory modules and shared modules (i.e., tissue-specific regulatory modules). As the validation process, we first performed literature survey to find the relationship of identified cohort-specific miRNAs and their corresponding cancer. We then applied experimentally supported validation of miRNA-mRNA interactions using miRTarBase database. Finally, to validate the performance of MSRFR model in distinguishing tissue- and cohort-specific regulatory modules, we explored the biological mechanism of identified key regulatory modules by checking their enriched pathways in PID and Hallmark collections. The overall result indicated that proposed model was able to effectively extract meaningful miRNA-mRNA regulatory modules and differentiate tissue- and cohort-specific modules.