Data-driven approaches to understand cancer-related phenotypes
In multicellular organisms, survival of the organism is favored over survival of individual cells. As such, normal cells are subject to limits on proliferation. In cancer, however, the accumulation of somatic, and sometimes germline, alterations converges to produce cells which are not subject to or can bypass normal growth restrictions, thus enabling the general cancer phenotype of dysregulated and excessive cell growth. Over the decades, we have made significant gains towards understanding how individual genetic alterations affect cell biology, and how these effects can converge to produce the cancer phenotype. Due to the cancer’s immense heterogeneity, however, we have yet to fully understand the whole-cell dynamic properties leading to the diversity of cancer sub-phenotypes that are observed in patients.The major theme of this dissertation is to interrogate many genetic backgrounds to characterize biological processes that are critical in the formation and/or maintenance of the general cancer phenotype. In chapter one, I performed a genome-wide screen to characterize the ultraviolet light-induced DNA damage response in the model organism Saccharomyces cerevisiae using a metric which reflects a time-dependent growth phenotype. This metric allowed me to identify a set of mitochondrial-associated genes involved in the response to ultraviolet-induced DNA damage. In chapter two, I used a context-aware deep learning model of therapeutic response to examine the cellular response to palbociclib, a selective inhibitor of the cyclin-dependent kinases four and six. I found that the response to palbociclib is governed by an array of distinct biological processes, and that patients and cell line samples are best stratified with the integration of all of these pathways. Overall, this body of work uses specific measures of context to further characterize biological processes critical to the development and maintenance of the cancer phenotype.