Molecular Drug Design Through Deep Reinforcement Learning

Deep reinforcement learning is an artificial intelligence technology that combines the perception of deep learning and the decision-making ability of reinforcement learning,with broad application prospects.In the field of medical research,the application of deep intensive learning in the discovery of new drugs needs further research and development.The nuclear export protein CRM1 is responsible for the nuclear transport of various tumor suppressor and cell cycle related proteins.In a variety of tumor cells,overexpression of CRM1 leads to the incorrect localization of tumor suppressor factors.By interfering with the CRM1-mediated transport mechanism,the occurrence and development of various human diseases related to cell proliferation and mislocalization of proteins can be effectively alleviated.In recent years,CRM1 has attracted great attention as a potential therapeutic target for tumors.Extranodal NK/T-cell lymphoma(ENKTL)is a subtype of non-Hodgkin’s lymphoma with high malignancy and poor prognosis.There is currently no effective treatment.Targeted therapy for tumors is characterized by high efficiency and safety.Targeted treatment of extranodal NK/T cell lymphoma is a promising research direction.In this study,artificial intelligence technology based on deep reinforcement learning was used to design a CRM1 small molecule inhibitor with better targeting and safety,providing new treatment options for extranodal NK/T cell lymphoma.The main conclusions of this study are as follows:(1)CRM1 targeting small molecule inhibitor LFS-1107 was designed by deep reinforcement learning.The interaction between LFS-1107 and CRM1 protein was detected by molecular docking and molecular dynamics simulation.The target of CRM1 protein is the cysteine residue located in its hydrophobic active pocket.Compared with sulforaphene,LFS-1107 showed better binding strength.The main amino acid residues that play a hydrophobic interaction between the small molecule inhibitors and CRM1 were different.(2)The selective killing of extracellular NK/T cell lymphoma by LFS-1107 was confirmed by cell viability assay,which had no effect on normal peripheral blood mononuclear lymphocytes(PBMC)and platelets.It was found by immunofluorescence experiments that LFS-1107 can inhibit the nuclear export of IκB-α by targeting CRM1.Western blotting and Elisa results demonstrated that LFS-1107 can down-regulate the NF-κB signaling pathway.Combined with the transcriptome experiments,it was found that LFS-1107 can up-regulate the apoptotic pathway and promote apoptosis in ENKTL cells.In this study,a more efficient CRM1 small molecule inhibitor LFS-1107 was designed by deep reinforcement learning.The inhibition of LFS-1107 on the proliferation of extranodal NK/T cell lymphoma was detected and the mechanism was explained.