Leveraging Cancer Genomics to Answer Clinically Motivated Questions: A Statistical Perspective

Abstract

Cancer is the number two cause of mortality in New Zealand. Up until the late 1990s there was a concerted effort by drug companies to develop 'blockbuster' therapies for the treatment of cancer, that is cancer therapies developed with a one-size-fits-all approach. This philosophy was upended by the arrival of patient-specific therapies developed from genomic-enabled cancer medicine, often referred to as Precision Oncology. Using genomics to determine what treatment an individual patient should receive as part of precision oncology created a new type of clinical test: a companion diagnostic. Companion diagnostics were either based on single genes or on multiple genes used together, in which case they were referred to as 'signatures'. The use of signatures in precision oncology is often based on gene expression and has been employed for a range of tasks including: estimating cancer grade, assessing immunological responses to a tumour and stratifying tumours into clinically relevant sub-types. While companion diagnostics have been transformative in oncology, the tumour biology that each test reflects is surprisingly poorly understood. This thesis explores the biologic foundations of cancer genomics with an emphasis on companion diagnostics and is broken into six chapters. The thesis begins by introducing in the fields of molecular biology, statistics, cancer biology and gene signatures. Then using novel analytical pipelines, the biological information reflected in commonly used cancer genomic signatures assessed. Here, previously obscured biology contained within each signature that was not obvious given the genes involved was identified. Next, an interactive gene-signature-focused web tool that allowed identification of signature components (principal and independent components) that are most associated with the Hallmarks of Cancer and prognostically important was created; this led to the generation of a novel pathway analysis pipeline to identify distinct time-dependent biological associations. The new pipeline was focused on breast cancer and its 'molecular subtypes'; these results suggest genomic information may be especially informative about the biological underpinnings of prognosis for patients with these subtypes of breast cancer Finally, a web tool was generated for interrogation of multi-omic data. This web tool allows researchers to rapidly identify and test hypotheses about the role of gene methylation, mutation, and copy number in determining tumour gene expression. Leveraging this tool, I suggest that DNA methylation and copy number aberration play largely mutually exclusive roles in driving oncogenic changes in gene expression in tumours. This analysis also suggests that the expression of genes linked to individual Hallmarks of Cancer, with the notable exception of immune response, is conserved between clinically different tumour types.