Two tier combinatorial structure to infer disease specific coexpression network

Discovering of biological mechanisms, which causes diseases at the molecular level, has been an important field of study in recent years. After the completion of the human genome project and the progress of DNA sequencing techniques, the relationships of molecules, which are related to cancer related biological processes, are revealed more quickly and easily. Gene coexpression networks are networks of molecular relationships formed by relation between genes which present similar patterns in samples with different phenotypes. Genes, which shows similar patterns on biological datasets, were found to have similar biological functions and involve in similar biological processes according to the studies in literature. In the thesis, gene coexpression networks were obtained by using different gene network inference algorithms on microarray gene expression, RNA-Seq and miRNA-target gene expression data. In the literature, there are limited number of studies that use different biological data sets and gene network inference algorithms in the same structure to construct a comprehensive and accurate gene coexpression networks for diseases. Our main target in thesis is to build up high-precision and comprehensive gene networks for breast and prostate cancer using different gene network inference algorithms and biological data sets together. According to this purpose, integration of gene coexpression networks has been achieved by forming a two-tier structure with basic integration methods such as intersection, simple majority voting and union. Gene expression networks, which are inferred by different gene network inference algorithms on the same data, are integrated in the first integration phase of the two-tier structure. Gene expression networks that are inferred from different biological datasets are integrated in the second integration phase of two- tier structure. The performance of the obtained gene association networks was evaluated according to their biological and topological properties. In addition to these two evaluation criteria, an overlap analysis was performed with the literature data. The usage of different gene network inference algorithms at integration phase slightly increases performance slightly. However, integration of gene coexpression networks that are infered on different biological datasets enhances the performances. We also developed a hash based association rule algorithm that infers potential gene coexpression networks of 152 diseases from miRNA- target gene data. We also present our algorithm as ARNetMiT R package.