Analysis of multivariate drug sensitivity dependence structure using copulas

Modeling sensitivity to anti-cancer drugs is a significant challenge in the area of systems medicine. Majority of current approaches generate a deterministic predictive model for each drug based on the genetic characterization of the cell lines and ignores the relationship between different drug sensitivities during model generation. In this article, we approach the problem of modeling the relationship between different drugs using probabilistic concept of copulas and generate the multivariate distribution of the drugs based on the marginal distributions of individual models and the estimated copula. We first illustrate using drug sensitivity databases that specific forms of copulas can be suitable for modeling the multivariate distribution of drug sensitivities. Subsequently, we show that parametric copulas estimated from training data can be utilized to increase the conditional sensitivity prediction accuracy of testing data as compared to prediction assuming independence between drug sensitivities.