MACHINE LEARNING BASED INTEGRATION OF miRNA AND mRNA PROFILES COMBINED WITH FEATURE GROUPING AND RANKING

Abstract

It is very important to understand the development and progression mechanisms of the diseases at the molecular level. Revealing the functional mechanisms that cause the disease not only contributes to the molecular diagnosis of the diseases, but also contributes to the development of the new treatment methods. Nowadays, due to the advances in technology, more molecular data can be obtained at cheaper costs, unlike in the past. Integrating these available data is essential to understand the molecular mechanisms of the diseases, especially the ones having complex formation and progression processes such as cancer. In this thesis, to correctly classify cancer patients and cancer free patients, two different bioinformatics tools (miRcorrNet and miRMUTINet) that integrate mRNA and microRNA data (two types of -omic data at the molecular level) have been developed. For 11 cancer types, mRNA and miRNA expression profiles of the samples were downloaded from The Cancer Genome Atlas. These two data types were integrated using both the Pearson Correlation Coefficient and the Mutual Information metrics. In our experiments using 100-fold Monte Carlo Cross Validation, for both tools, 99% Area Under the Curve score have been obtained. The developed tools have also been tested using independent dataset. For biological validation purposes, for each cancer type, functional enrichment analysis is conducted on the identified list of significant miRNAs and genes. Additionally, for each cancer type, the identified mRNAs and miRNAs were subject to literature validation and the findings were noteworthy