Distinguished Speaker Series: Models and algorithms to identify alterations associated with drug response sensitivity in cancer

Abstract:  

The identification of alterations and mutations associated with drug sensitivity responses is key to predictive genomics - the personalized prediction of whether a patient will respond positively or negatively to a drug.   The recent availability of data that contain information on the gene profiles of patients that have been treated with a specific drug as well as the patients' individual responses to the drug makes this identification possible.  Using this data it is desired to identify those specific alterations that affect the response (positive or negative) to a certain drug.

We describe models and algorithms to identify alterations with complementary functional associations in cancer by modeling the problem as a maximum set coverage problem that requires that the alterations are mutual exclusive and/or form a connected associated subnetwork.  A generalization of these models that allows for positive or "negative" coverage is used to identify alterations associated with specific drug responses. The methods employed include integer programming, heuristics and approximation algorithms.   These maximum coverage methods can be used as feature selection approaches in machine learning for predicting drug response.  A preliminary exploration indicates that this leads to significant improvement in the accuracy of machine learning techniques as compared with known feature reduction methods.

(joint with Rebecca Sarto Basso, Fabio Vandin, Teresa Przetycka and Yoo-Ah Kim).

Host: Prof. Ron Shamir, Computer Science School, TAU.