Graph-based Support Vector Machines for Patient Response Prediction Using Pathway and Gene Expression Data

Over the past decade, multiple function genomic datasets studying chromosomal aberrations and their downstream implications on gene expression have accumulated across a variety of cancer types. With the majority being paired copy number/gene expression profiles originating from the same patient groups, this time frame has also induced a wealth of integrative attempts in hope that the concurrent analysis between both genomic structures will result in optimized downstream results. Borrowing the concept, this dissertation presents a novel contribution to the development of statistical methodology for integrating copy number and gene expression data for purposes of predicting treatment response in multiple myeloma patients.;This dissertation is structured in three complimentary sections. The first reviews the methods currently available for integrative purposes between gene expression and copy number data. Specifically this includes the conceptual evolution of these workflows, approaches used amongst varying methods, endpoints targeted for downstream analysis, and biological milestones achieved through such efforts. The focus here is to highlight the accomplishments and potential areas for improvement. A key takeaway message is the lack of integrative attempts in the field of response prediction.;The second section consequently introduces a new integrative approach for response prediction. This section is furthermore split into two subsections where the first describes the motivation, intuition, theoretical developments, and simulation/application results with respect to the proposal; while the second describes an extension to include copy number data. Note that since the approach introduced in the initial subsection only utilizes the gene expression data, it will therefore require the latter argument to complete its integrative design.;The final section then concludes the dissertation by discussing future steps in data integration and how these innovations can potentially lead to more efficient and robust response prediction models.