De novo design of dual-target JAK2, SMO inhibitors based on deep reinforcement learning, molecular docking and molecular dynamics simulations

Abstract

Triple-negative breast cancer (TNBC) and HER2-positive breast cancer are particularly aggressive and the effectiveness of current therapies for them is limited. TNBC lacks effective therapies and HER2-positive cancer is often resistant to HER2-targeted drugs after an initial response. The recent studies have demonstrated that the combination of JAK2 inhibitors and SMO inhibitors can effectively inhibit the growth and metastasis of TNBC and HER2-positive drug resistant breast cancer cells. In this study, deep reinforcement learning was used to learn the characteristics of existing small molecule inhibitors of JAK2 and SMO, and to generate a novel library of small molecule compounds that may be able to inhibit both JAK2 and SMO. Subsequently, the molecule library was screened by molecular docking and a total of 7 compounds were selected out as dual inhibitors of JAK2 and SMO. Molecular dynamics simulations and binding free energies showed that the top three compounds stably bound to both JAK2 and SMO proteins. The binding free energies and hydrogen bond occupancy of key amino acids indicate that A8976 and A10625 has good properties and could be a potential dual-target inhibitor of JAK2 and SMO.

Keywords: Deep reinforcement learning; Dual-target inhibitor; Dynamic simulation; TNBC.