UC San Diego

The genome of a cancer cell is replete with somatic mutations imprinted by the activities of different endogenous and exogenous processes. Each mutational process exhibits a characteristic pattern of mutations, termed mutational signature. Prior work has shown that mutational signatures can be deciphered from a set of cancer genomes, thus, providing insight into the mutagenic processes that have been operative throughout the lineage of the cancer cell. Analysis of mutational signatures has had three major applications: (i) leveraging mutational signatures to identify environmental mutagens that cause cancer, thus, providing opportunities for developing cancer prevention strategies; (ii) using mutational signatures to better understand the biological mechanisms of DNA damage and repair processes; (iii) utilizing mutational signatures of failed DNA repair as biomarkers for targeted cancer treatment. However, the universal deployment of mutational signatures has been limited mainly by a reliance on whole-genome sequencing and downstream expert interpretation.

In this dissertation, we first develop three novel computational frameworks for exploring mutational signatures from large cohorts of cancer. We apply these approaches in a pan-cancer analysis to elucidate the mutational processes giving rise to clustered mutational events encompassing a plethora of operative endogenous and exogenous processes. Comprehensive characterization of these events reveals an enrichment within known driver genes. Importantly, clustered driver mutations are detectable from standard-of-care diagnostic tests and can serve as prognostic biomarkers for the overall survival of a cancer patient. Further, we introduce a novel form of oncogenesis, termed kyklonas, indicative of a repeated hypermutation of extrachromosomal circular DNA driven by the innate immune system.

Lastly, we propose an alternative sequencing-independent and cost-effective method for detecting mutational signatures by applying a deep learning approach to digital images of histopathological cancer slides. We demonstrate both the ability of this novel approach for detecting homologous recombination deficiency within breast and ovarian cancers as well as its clinical utility for predicting sensitivity to platinum treatment in individual cancer patients.