RIPK1 inhibitor generated via deep learning

A small molecule inhibitor of the kinase RIPK1 designed using a generative deep learning model could help treat inflammatory diseases including IBD.

A distribution-learning conditional recurrent neural network trained on 1,030 known RIPK1 inhibitors yielded 79,323 molecules, and subsequent virtual screening, molecular docking, chemical synthesis and in vitro assays identified a compound that inhibited RIPK1 with an IC50 of 35 nM. The compound had triazolo[1,5-a] pyridine and indole moieties that bound RIPK1's ATP-binding pocket, and a terminal benzene ring that bound the kinase's allosteric site...