Wail Ba-Alawi 1,2, Benjamin Haibe-Kains 1,2
1 Princess Margaret Cancer Centre, University Health Network
2 Department of Medical Biophysics, University of Toronto
Abstract
A key step in personalized cancer medicine is to identify biomarkers that can predict how different cancer cells will respond to a drug treatment. This is a challenging problem given the lack of clinical trials where both tumor molecular profiles and patient therapy outcome are available. In order to alleviate this problem, we leverage big pharmacogenomic data to build an innovative computational approach aiming at finding genomic and molecular features predictive of drug response across the different cancer types. To identify robust and actionable biomarkers, we fit logical models to produce a set of small and interpretable rules to predict drug response. Our results show that we can find good predictors for more than 60% of the drugs in GDSC data set. We were able to validate around 70% of these multivariate biomarkers in external data sets.