Jan 23
Math/Stats Colloquium: Jaime Davila, St. Olaf College
Title: Unsupervised Learning in Mutational Signatures Elicits Cancer Diagnostic Clues
Abstract: One of the hallmarks of cancer is the presence of somatic mutations, which are changes in the DNA of the tumor cell. The pattern in which such mutation occurs is called a mutational signature. Particular mutational signatures are caused by specific processes, for example, UV light causes CC to TT mutations, while tobacco smoking is characterized by the presence of C to A mutations. During this talk I will present my efforts in using mutational signatures to determine the tissue of origin in a clinical cohort with metastatic tumors of unknown primary. The approach I used leverages unsupervised learning, a branch from machine learning that aims to discover patterns in unlabeled data. In particular, I will describe my favorite unsupervised learning technique, non-negative matrix factorization, and describe how to use it on mutational signatures from clinical cancer cases to establish tissue of origin and provide diagnostic clues.