Methods for Characterizing the Activity of Mutational Processes in Cancer
The cancer sequencing efforts of the past decade have revealed signatures of the mutational processes shaping cancer genomes. These mutational signatures provide a window into a tumor’s functional state and history, and thus provide new opportunities for identifying the mutations driving an individual’s cancer and for personalized medicine. While researchers have now collected >30 validated mutational signatures, challenges remain for understanding the patterns of mutational signature activity. One such challenge is in characterizing signature etiology: many signatures have unknown etiology, while some similar signatures have different etiologies.
In this talk, we will present two probabilistic methods that begin to address these challenges. Inspired by research from natural language processing, the first method, TCSM, models mutational signature activity per tumor conditioned on observed metadata about the patient. We will show that TCSM outperforms standard methods at inferring mutational signature activity and for inferring clinically relevant DNA damage repair deficiencies in breast cancer. Next, we will present SigMa, the first model of mutational signature activity to account for sequence dependencies among clustered mutations. We use these inferred dependencies and associations with other genomic factors to reveal new insights into signature etiology. Finally, we will conclude by presenting ongoing work on ExploSig, a family of tools to enable biologists and data scientists to explore mutational signatures datasets in the browser and in interactive notebooks.
Course Material 1: leiserson-nci-seminar-2019-1.pptx
Course Material 2: Max-Leiserson-BTEP.zip
Log in to access student content for this class.