CANCER GENOME BIOMARKER DISCOVERY WITH INTEGRATED MOLECULAR DATA AND SYSTEMS BIOLOGY-ORIENTED APPROACHES

Abstract

The discovery of sequence variants causal to patient survival outcomes had been one of the major aims in cancer genomics. However, genetic heterogeneity is ubiquitously observed in tumors and has led to difficulties in detecting significant signals in the population study setting. In this thesis, I propose a novel segmentation-based approach called Gene-to-Protein-to-Disease (GPD), which facilitates the detection of reproducible prognostic signals across multiple patients by aggregating genetic alternations to protein units. These units are derived from information on protein domains and modification sites through which proteins carry out their functions. A part of the thesis is devoted to creating an in-silico method called PTMscape for predicting proteome-wide modification sites, which leads to the more comprehensive annotation of protein-coding regions for GPD analysis. I demonstrate the proposed computational approaches in the Pan-Cancer Atlas data obtained from ~10,000 genomes across 33 cancer types and present the discovery of 247 survival-associated protein units.