Proteome-scale detection of drug-target interactions using correlations in transcriptomic perturbations

Systems biologyseeks to understand how normal and disease protein networks respond when specific interactions are disrupted. A first step towards this goal is identifying the molecular target(s) of bioactive compounds. Here, we hypothesize that inhibitory drugs should produce network- level effectssimilar to silencing the inhibited gene and show that drug-protein interactions are encoded in mRNA expression profile correlations. We use machine learning to classify correlations between drug- and knockdown-induced expression signatures and enrich our predictions through a structure-based screen. Interactions manifest both as direct correlations between drug and target knockdowns, and as indirect correlations with up/downstream knockdowns. Cross-validation on 152 FDA- approved drugsand 3104 potential targets achieved top 10/100 prediction accuracies of 26/41%. We apply our method to 1680 bioactive compoundsand experimentally validate five previously unknown interactions. Our pipeline can accelerate drug discovery by matching existing compounds to new therapeutic targets while informing on network and multi-target effects.