Theoretical Design And Screening Of Severe Acute Respiratory Syndrome Coronavirus 2 Main Protease Inhibitors

The Coronavirus disease 2019 has caused significant harm to people’s health,and developing inhibitors against the severe acute respiratory syndrome coronavirus 2(SARS-Co V-2)is an effective way to deal with the pandemic.Since the SARS-Co V-2Main protease(Mpro)plays an essential role in the virus replication process and has no homologous target in the human body,it is an ideal drug target.The development of computer science has driven the advancement of cheminformatics,including computeraided drug design and artificial intelligence drug design,which can help people design and screen drug molecules more efficiently.This thesis employs molecular docking and molecular dynamics simulations to study the interaction mechanisms between some molecules and the SARS-Co V-2 Mpro.By combining similarity search and quantitative structure-activity relationship screening,several inhibitors were identified.New inhibitor molecules were generated using molecular generation methods.The main research content of the thesis is as follows:Chapter 1: The section introduces the SARS-Co V-2,its main protease,and related inhibitors,as well as providing a brief overview of computer-aided drug design and artificial intelligence drug design.It also summarizes the application of these methods in the design of drugs targeting the SARS-Co V-2 and the investigation of their related mechanisms.Chapter 2: A representative class of peptidomimetic molecules,11 a and PF-07321332,are effective inhibitors of the SARS-Co V-2 Mpro.In this chapter,molecular dynamics simulations were used to explore the interaction mechanisms of 11 a and PF-07321332 with the SARS-Co V-2 Mpro.The results show that the P1 and P1’ fragments of such molecules play an essential role in the binding process,while the P2 and P3 fragments have optimization potential.Therefore,we selected some fragment replacements for the original P2 and P3 fragments targeting the main protease inhibitors of SARS-Co V and SARS-Co V-2.After obtaining new molecules,molecular docking and molecular dynamics simulations were conducted for further screening,resulting in two potential inhibitors.The potential inhibitors were further confirmed through ADMET property predictions.Chapter 3: The similarity principle is an essential foundation for ligand-based drug design.Depending on whether the similarity has global or local characteristics,molecular similarity can be divided into global similarity and local similarity.In this chapter,we used similarity search and quantitative structure-activity relationship models to screen for SARS-Co V-2 Mpro inhibitors,which can comprehensively consider both global and local similarities.First,we used machine learning algorithms to establish a quantitative structure-activity relationship model based on the existing dataset of active SARS-Co V-2 inhibitors.The best model had a ROC-AUC of 0.9110 in the test set,indicating that it could effectively distinguish between highly active and less active inhibitors.Then,using the SHAP method,we analyzed and explained the established model,and found that the presence of the pyridine ring was beneficial for inhibitor activity.In addition,we screened molecules using the similarity search method and combined them with the previously established quantitative structure-activity relationship to obtain 11 possible inhibitor molecules.Finally,MM/GBSA calculations after molecular docking indicated that the screened molecules might be the potential Mpro inhibitors.Chapter 4: Virtual screening is generally performed by molecular docking or trained models to screen potential active molecules from a large database.However,the molecules in existing databases only cover a portion of the chemical space.Molecular generation methods can be used to explore a broader range of molecules in the chemical space.In this chapter,we trained a Variational Autoencoder(VAE)model to generate molecules around the chemical space of active inhibitors.The Long Short Term Memory(LSTM)neural network was employed to process molecules represented by SMILES.We trained the model in two steps: first,training the model on the Ch EMBL dataset,and then fine-tuning it on the active inhibitor dataset.During the molecular generation phase,we added random perturbations to the encoder’s output to achieve sampling,enabling the model to generate molecules around the chemical space of active inhibitors.Computational results showed that the property distribution of the generated molecules was close to that of the original inhibitors molecules.Finally,the generated molecules were subjected to ADMET screening,synthesizability scoring,and molecular docking simulations.The results indicated that some of the molecules generated by the model had high docking scores and the potential to become new inhibitor molecules.