Design and discovery of monopolar spindle kinase 1 (MPS1/TTK) inhibitors by computational approaches

Monopolar spindle kinase 1 (MPS1, also called TTK) is an attractive target for the treatment of cancers. Five MPS1 inhibitors have entered the clinical trials, but four of them were discontinued; thus, it is necessary to develop MPS1 inhibitors with novel scaffolds. In the present work, several computational tools were built to design MPS1 inhibitors. The deep recurrent neural network was used for generating potential highly active MPS1 inhibitors. The deep neural network was used to build a ligand-based consensus model for distinguishing the highly and weakly active MPS1 inhibitors. Five co-crystal structures were used to develop the consensus docking score for distinguishing actives and decoys. The ligand-based consensus model and the consensus docking score were used to evaluate the generated molecules, and finally, two scaffolds, which were less reported as MPS1 inhibitors previously, were selected. A total of 15 compounds with the two scaffolds were synthesized and tested by in vitro enzymatic inhibition assays. Five compounds had sub-micromolar to low micromolar in vitro potencies, and the most active compound was 10 with an IC₅₀ of 556 nM. The binding modes of the new compounds were revealed by molecular dynamic simulations. We believe that the computational strategies in the present work were helpful for discovering new potential scaffolds.