Date of Award

Spring 2022

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Computational Biology and Bioinformatics

First Advisor

Gerstein, Mark

Abstract

Advances in genomic sequencing technologies opened the door for a wider study of cancer etiology. By analyzing datasets with thousands of exomes (or genomes), researchers gained a better understanding of the genomic alterations that confer a selective advantage towards cancerous growth. A predominant narrative in the field has been based on a dichotomy of alterations that confer a strong selective advantage, called cancer drivers, and the bulk of other alterations assumed to have a neutral effect, called passengers. Yet, a series of studies questioned this narrative and assigned potential roles to passengers, be it in terms of facilitating tumorigenesis or countering the effect of drivers. Consequently, the passenger mutational landscape received a higher level of attention in attempt to prioritize the possible effects of its alterations and to identify new therapeutic targets. In this dissertation, we introduce interpretable network approaches to the study of genomic variation in cancer. We rely on two types of networks, namely functional biological networks and artificial neural nets. In the first chapter, we describe a propagation method that prioritizes 230 infrequently mutated genes with respect to their potential contribution to cancer development. In the second chapter, we further transcend the driver-passenger dichotomy and demonstrate a gradient of cancer relevance across human genes. In the last two chapters, we present methods that simplify neural network models to render them more interpretable with a focus on functional genomic applications in cancer and beyond.

Download
COinS