Abstract The outbreak of coronavirus disease 2019 (COVID-19) SARS-CoV-2 has caused widespread panic in the world and has mutated at an extremely rapid rate and thus there is an urgent need for the development of COVID-19 inhibitors. In this study, we used a de novo design method, which integrates a recurrent neural network, reinforcement learning and molecular docking to generate inhibitors of SARS-CoV-2 main protease. Approximately 30,000 molecules were generated after a 120h generation process, and multiple physicochemical filters and molecular docking scores were used for further screening. Finally, five molecules were selected as drug candidates, and their binding stability was verified by molecular dynamics simulation and binding free energy analysis. The results showed that these molecules could be used as candidates for further generation and testing against SARS-CoV-2. Besides, a pharmacophore model based on superior molecules was constructed to provide a reference for subsequent drug screening.
Recent outbreaks of coronavirus have brought serious challenges to public health around the world, and it is essential to find effective treatments. In this study, the 3C-like proteinase (3CLpro) of SARS-CoV-2 (severe acute respiratory syndrome coronavirus 2) has been considered as an important drug target because of its role in viral replication. We initially optimized 251 compounds at the PM7 level of theory for docking with 3CLpro, and then we selected the top 12 compounds for further optimization with the B3LYP-D3/6-311G** method and obtained the top four compounds by further molecular docking. Quantum chemistry calculations were performed to predict molecular properties, such as the electrostatic potential and some CDFT descriptors. We also performed molecular dynamics simulations and free energy calculations to determine the relative stability of the selected four potential compounds. We have identified key residues controlling the 3CLpro/ligand binding from per-residue based decomposition of the binding free energy. Convincingly, the comprehensive results support the conclusion that the compounds have the potential to become a candidate for anti-coronavirus treatment.
Since the main protease (Mpro) is crucial for the COVID-19 virus replication and transcription, searching for Mpro inhibitors is one possible treatment option. In our study, 258 small molecules were collected from lung-related herbal medicines, and their structures were optimized with the B3LYP-D3/6-31G* method. After the molecular docking with Mpro, we selected the top 20 compounds for the further geometry optimization with the larger basis sets. After the further molecular docking, the top eight compounds were screened out. Then we performed molecular dynamics simulations and binding free energy calculations to determine stability of the complexes. Our results show that mulberrofuran G, Xambioona, and kuwanon D can bind Mpro well. In quantum chemistry studies, such as ESP and CDFT analyses, the compounds properties are predicted. Additionally, the drug-likeness analyses and ADME studies on these three candidate compounds verified that all of them conform to Libinski’s rule and may be drug-like compounds.
Although the COVID-19 pandemic has been brought under control to some extent globally, there is still debate in the industry about the feasibility of using artificial intelligence (AI) to generate COVID small-molecule inhibitors. In this study, we explored the feasibility of using AI to design effective inhibitors of COVID-19. By combining a generative model with reinforcement learning and molecular docking, we designed small-molecule inhibitors targeting the COVID-19 3CLpro enzyme. After screening based on molecular docking scores and physicochemical properties, we obtained five candidate inhibitors. Furthermore, theoretical calculations confirmed that these candidate inhibitors have significant binding stability with COVID-19 3CLpro, comparable to or better than existing COVID-19 inhibitors. Additionally, through ligand-based pharmacophore model screening, we validated the effectiveness of the generative model, demonstrating the potential value of AI in drug design.
'/%3e%3cuse xlink:href='%23a' transform='rotate(23.036 .093 25.536)'/%3e%3cuse xlink:href='%23a' transform='rotate(45.87 1.273 16.18)'/%3e%3cuse xlink:href='%23a' transform='rotate(69.945 .996 12.078)'/%3e%3cuse xlink:href='%23a' transform='rotate(20.66 -19.689 31.932)'/%3e%3c/svg%3e)