BAYESIAN INTEGRATIVE ANALYSIS OF OMICS DATA

Technological innovations have produced large multi-modal datasets that range in multiplatform genomic data, pathway data, proteomic data, imaging data and clinical data. Integrative analysis of such data sets have potentiality in revealing important biological and clinical insights into complex diseases like cancer. This dissertation focuses on Bayesian methodology establishment in integrative analysis of radiogenomics and pathway driver detection applied in cancer applications. We initially present Radio-iBAG that utilizes Bayesian approaches in analyzing radiological imaging and multi-platform genomic data, which we establish a multi-scale Bayesian hierarchical model that simultaneously identifies genomic and radiomic, i.e., radiology-based imaging markers, along with the latent associations between these two modalities, and to detect the overall prognostic relevance of the combined markers. Our method is motivated by and applied to The Cancer Genome Atlas glioblastoma multiforme data set, wherein it identifies important magnetic resonance imaging features and the associated genomic platforms that are also significantly related with patient survival times. For another aspect of integrative analysis, we then present pathDrive that aims to detect key genetic and epigenetic upstream drivers that influence pathway activity. The method is applied into colorectal cancer incorporated with its four molecular subtypes. For each of the pathways that significantly differentiates subgroups, we detect important genomic drivers that can be viewed as “switches” for the pathway activity. To extend the analysis, finally, we develop proteomic based pathway driver analysis for multiple cancer types wherein we simultaneously detect genomic upstream factors that influence a specific pathway for each cancer type within the cancer group. With Bayesian hierarchical model, we detect signals borrowing strength from common cancer type to rare cancer type, and simultaneously estimate their selection similarity. Through simulation study, our method is demonstrated in providing many advantages, including increased power and lower false discovery rates. We then apply the method into the analysis of multiple cancer groups, wherein we detect key genomic upstream drivers with proper biological interpretation. The overall framework and methodologies established in this dissertation illustrate further investigation in the field of integrative analysis of omics data, provide more comprehensive insight into biological mechanisms and processes, cancer development and progression