correlation between gCSI and CTRP data sets is significantly increased using GR AOC rather than AUC as a response metric (P = 1.3 × 10 -3 , Student's t-test; Supplementary Fig. 4).And third, efficacy as measured by GR max and potency as measured by GR 50 differ at a biological level, carry complementary information (low mutual information), and are associated with largely non-overlapping genetic alterations. In principle, variation in potency and efficacy can be captured by integrating across dose-response curves (GR AOC ), but we find that information content is maximized if GR 50 and GR max are considered independently.Because the ultimate purpose of antineoplastic drugs is to kill cancer cells 13 , and high potency is no guarantee of good efficacy, we propose that the best drugs and most important pharmacogenomic associations are not those associated with low IC 50 values, but rather those that result in the most negative GR value at clinically relevant drug concentrations (e.g., C max ). Relating in vitro measures of drug sensitivity to in vivo responses remains challenging 17 , but for this to have any chance of success it is essential that in vitro data are as informative and reproducible as possible. Editor's note: This article has been peer-reviewed.