The hunt for causal mutations driving cancer has been driven by the promise of understanding the basic biology of tumors, forecasting patient outcome, and finding druggable targets. Most approaches to finding relevant cancer genes seek those that are highly recurrently altered. But problems with this principle include the high prevalence of passenger mutations; subtypes of cancer with different advantageous mutations; the long lists of uninterpretable candidate genes that are produced; and the fact that not one mutation but combinations of genetic alterations lead to cancer. In this dissertation, I will consider different sources of evidence to prioritize genes genetically altered in cancer, and to understand how these lead to tumor development, with a particular focus on melanoma and glioblastoma. I develop an unsupervised information theoretic method that addresses the combinatorial nature of genetic alterations in cancer. Then, I consider that combinations of co-occurring Mendelian disease and cancer can provide insights into oncogenesis.  Using results from population-scale Electronic Health Records along with information on causal germline mutations in Mendelian disease and somatic mutations in cancer, I show that comorbid Mendelian disease and cancer are more likely to involve similar genes, putting forward comorbidity as a source for novel cancer driver genes.