Oncology drugs are only effective in a small proportion of cancer patients. To make things worse, our current ability to identify these responsive patients before treatment is still very limited. Thus, there is a pressing need to discover response markers for marketed and research oncology drugs in order to improve patient survival, reduce healthcare costs and enhance success rates in clinical trials. Screening these drugs against a large panel of cancer cell lines has been recently employed to discover new genomic markers of in vitro drug response. However, while the identification of these markers among several thousands of candidate drug-gene associations is error-prone, an appraisal of the effectiveness of such detection task is currently lacking. Here we propose a new approach that directly measures the discrimination power of a drug-gene association by posing each of these associations as a binary classification problem. The application of this methodology has led to the identification of 232 new genomic markers distributed across 81% of the analysed drugs, including 8 drugs without previously known markers, which were missed by the methodology initially applied to the Genomics of Drug Sensitivity in Cancer (GDSC) dataset.