Multivariate Bayesian structured variable selection for pharmacogenomic studies

Precision cancer medicine aims to determine the optimal treatment for each patient. In-vitro cancer drug sensitivity screens combined with multi-omics characterization of the cancer cells have become an important tool to achieve this aim. Analyzing such pharmacogenomic studies requires flexible and efficient joint statistical models for associating drug sensitivity with high-dimensional multi-omics data. We propose a multivariate Bayesian structured variable selection model for sparse identification of omics features associated with multiple correlated drug responses. Since many anti-cancer drugs are designed for specific molecular targets, our approach makes use of known structure between responses and predictors, e.g. molecular pathways and related omics features targeted by specific drugs, via a Markov random field (MRF) prior for the latent indicator variables of the coefficients in sparse seemingly unrelated regression. The structure information included in the MRF prior can improve the model performance, i.e. variable selection and response prediction, compared to other common priors. In addition, we employ random effects to capture heterogeneity between cancer types in a pan-cancer setting. The proposed approach is validated by simulation studies and applied to the Genomics of Drug Sensitivity in Cancer data, which includes pharmacological profiling and multi-omics characterization of a large set of heterogeneous cell lines.