Fig. 3

Global drug module analysis. a Number of genes in positively and negatively correlated modules (PCMs and NCMs) (left). Proportion of genes in the modules with respect to PCGs and NCGs (i.e., full signature). b Distance between drug targets and PCMs/NCMs (purple cumulative distribution). Results are compared to random proteins from the STRING interactome (red), proteins sampled from the “druggable proteome” (Target Central Resource Database) (green), and proteins sampled from the pharmacological targets in DrugBank (blue). c Network-based distances between drug classes. The bigger the bubble, the closer the distance between drug classes. Drug classes are sorted by specificity in their proximity measures. Please note “distant” values in the diagonal are possible due to differences in drug modules belonging to the same class. The network-based distance calculation is detailed in the “Methods” section. d Predictive performances (AUROC) of the drug modules evaluated in the CTRP panel (top 25, 50, and 100 sensitive CCLs). Blue distributions correspond to results using unique CCLs (i.e., not shared with the GDSC panel). e Illustrative ROC curves for Daporinad (FMK866), Vorinostat, I-BET-762, and Docetaxel