Design And Study Of Small Molecule Inhibitors Of SARS-CoV-2 Mpro,PLpro

Since December 2019,the new coronavirus has been spreading rapidly around the world,killing 6 million people and affecting the livelihoods of 7 billion people worldwide up to now.Further research into disease pathogenesis in recent years has shown that single-target drugs often have limitations in the treatment of complex diseases.In this study,the two most critical proteins in the replication of SARS-Co V-2 virus,the main protease-like protease（Mpro）and papain-like protease（PLpro）proteins,which are highly conserved in structure and function,were targeted to reduce the risk of drug resistance development to a certain extent.Deep reinforcement learning has a unique advantage in balancing the strength of action between drug targets and can better design multi-target drug molecules.In this study,deep reinforcement learning was used to generate a database of specific small molecules,and the generated molecules were screened by target prediction activity,drug-like properties,and synthetic feasibility.Then,the generated molecules were screened and evaluated again using traditional drug design methods such as molecular docking,with the expectation of obtaining lead compounds with novel structures and good activities.This study will give the novel SARS-Co V-2Mpro/PLpro dual target covalent inhibitors,provide the foundation for small molecule drugs for the treatment of COVID-19 and provide the theoretical basis for the prevention and treatment of possible future epidemic coronaviruses.The main contents are as follows.1.We collected small molecule inhibitors of SARS-Co V-1 and SARS-Co V-2Mpro/PLpro reported in the literature and used deep reinforcement learning to learn the features of these small molecules and construct a library of small molecule compounds,where the generative model was a recurrent neural network（RNN）model and the discriminative model was a random forest（RF）model.The generative model was evaluated using the validity of the generated SMILES characters and achieved 98%validity after 200 epochs,indicating that the generative model could generate the correct chemical structure and the model was fine-tuned using the collected active molecules;the discriminative model was evaluated using the AUC curve and the area under the curve for the Mpro and PLpro discriminative models was 0.867 and 0.876,respectively.This indicates that the discriminative models have a good ability to distinguish between active and inactive compounds.Finally,a total of 4428 small molecules were generated,most of which contained structures such as benzene rings,naphthalene rings and pyrrolidone.The PCA dimensionality reduction revealed a high degree of chemical spatial overlap between the generated molecules and the known active molecules,indicating that the generated molecules learned the characteristics of the active molecules.2.For more powerful inhibition of SARS-Co V-2 virus,the generated compounds were virtually screened using covalent docking to select compounds that could covalently bind to the cysteines of both targets and the MMGBSA score was calculated.A total of two covalent docking processes were performed,the first using a virtual screening model to dock all previously obtained molecules and the top 20 compounds were selected for the second covalent docking using the pose prediction model and the MMGBSA score was calculated.Afterwards,three compounds A3175,A3659,and A3777 were selected based on the cdock affinity score,MMGBSA score,and compound interactions with key amino acid residues.All three compounds interacted with His41,Cys145,Glu166 of Mpro and Cys111 of PLpro,which are similar to known Mpro or PLpro inhibitors,indicating their potential as antiviral agents.3.The selected compounds were analyzed by molecular dynamics simulations for100 ns to study the stability of protein-ligand binding and to analyze the interaction of the compounds with the target protein during the dynamics simulations.The RMSD values of the complex systems were stable within 3 （?）,indicating that the compounds were stable in their binding to the protein.Also,high frequency interactions of the compounds with key amino acid residues of Mpro were observed during dynamics simulations and interactions of the compounds with the BL2 loop of PLpro were observed,suggesting that the three compounds have the potential to be dual target inhibitors of SARS-Co V-2 Mpro and PLpro.