One of the main challenges in precision medicine consists of developing predictors of drug response to select the most beneficial therapy for each individual patient. In this context, preclinical models are crucial to study the association between molecular features of tumor cells and response to chemical perturbations. However, only few predictors have been successfully translated to clinical settings. Such a low success rate is due not only to the complexity of the mechanisms underlying anticancer drug response, but also to the lack of robustness of the predictors developed in preclinical settings. To address this issue we developed PharmacoGx, a computational platform enabling meta-analysis of large-scale drug screenings of in vitro and in vivo model systems, and PharmacoDB (pharmacodb.pmgenomics.ca), a web-application enabling quick access to a large compendium of pharmacogenomics datasets. In this presentation I will show how we used our new platforms to (i) identify pathways whose genomic and transcriptomic patterns are preserved between model systems and patient tumors, and (ii) develop univariate and multivariate predictors of drug response that can be validated in clinical trial data.