Advancing Variant Effect Prediction Beyond Protein-Level by Incorporating Systems-Level Architecture

Cancer is a complex disease that harbors substantial genetic heterogeneity. Recent advances in sequencing technologies revealed large numbers of somatic mutations across human tumors. However only a small proportion of these mutations are expected to contribute to tumor growth and progression, making determining functional mutations an important challenge in cancer genomics. Missense variants are particularly difficult to understand as they change only a single amino acid in a protein sequence yet can have large and varied effects on protein activity. Numerous tools have been developed to identify missense variants with putative disease consequences from protein sequence and structure. However, biological function arises through higher order interactions among proteins and molecules within cells, and diseases are often associated with perturbations to protein interactions. Different perturbations can result in different phenotypes, and the level of impact caused by mutations to the underlying molecular interaction network may determine the likelihood of generating a disease phenotype. Thus, in this dissertation, I aim to incorporate systems-level architecture to bridge the gap between genotype and phenotype in cancer by exploring different network-based strategies to study the impact of variants on biological systems. I first integrated protein structure and network information to design variant features that capture orthogonal information to classical amino acid features and showed their potential to improve variant classification within a machine-learning framework. Next, I investigated how patterns of network rewiring of mutations on cancer genes can be informative for unearthing different selective oncogenic pressures. Finally, I examined transcriptomic effects of perturbation of distinct protein interactions as a way to better define the landscape of prospective phenotypes reachable by individual amino acid substitutions. Overall, this body of work demonstrates that variant effect interpretation can be significantly improved by incorporating information about the role of proteins and their molecular interactions within biological systems