Fig. 1From: Deep learning-driven prediction of drug mechanism of action from large-scale chemical-genetic interaction profilesOverview of the study design. The CGIPs of M. tuberculosis consist of growth inhibitions (in Z-scores) of ~ 50,000 chemicals against 152 M. tuberculosis mutant strains (hypomorphs). The gene-level clustering was first achieved through the following processes: homology search in BLAST, gene semantic similarity computation, and cluster identification using dynamic tree cut. Using the clustered data, we trained a directed message passing network, which learned a molecular graph for each compound from the molecular features generated by RDKit. Next, we measured the performance on the test set and applied the model on several curated chemicals for further evaluationBack to article page