Machine learning approaches to personalized medicine and genomics


Gerald Quon (MIT)
Thursday, September 24, 2015
4:00 PM – 5:15 PM
Shanahan Auditorium

One of the principal challenges of medicine is to find robust biomarkers, or predictors, of patient response to therapies and disease outcome, in order to better assign personalized treatments to patients. While the “Big Data” revolution in biology has enabled clinicians to measure millions of candidate biomarkers in patients, our ability to find robust predictors of treatment efficacy and disease outcome has lagged, in large part due to heterogeneity in patient cohorts, data types and measurement instruments. Fortunately, the field of machine learning has made tremendous progress in terms of developing mixture models that efficiently handle different types of heterogeneity.

In this talk, I will discuss how probabilistic topic models, originally designed to solve problems in information retrieval and nature language processing, are readily adapted to address problems in data heterogeneity in the clinic, and illustrate how they have been used to improve our ability to predict outcome of cancer patients.

Gerald Quon is an assistant professor in the Department of Molecular and Cellular Biology at UC Davis. He obtained his B. Math in Computer Science at the University of Waterloo, his Ph.D. in Computer Science from the University of Toronto, and completed postdoctoral research training at MIT.