Development of a deep learning-based quantitative structure-activity relationship model to identify potential inhibitors against the 3C-like protease of SARS-CoV-2

Background: In the recent COVID-19 pandemic, SARS-CoV-2 infection spread worldwide. The 3C-like protease (3CLpro) is a promising drug target for SARS-CoV-2. Results: We constructed a deep learning-based convolutional neural network-quantitative structure-activity relationship (CNN-QSAR) model and deployed it on various databases to predict the biological activity of 3CLpro inhibitors. Subsequently, molecular docking analysis, molecular dynamics simulations and binding free energy calculations were performed to validate the predicted inhibitory activity against 3CLpro of SARS-CoV-2. The model showed mean squared error = 0.114, mean absolute error = 0.24 and predicted R-2 = 0.84 for the test dataset. Diosmin showed good binding affinity and stability over the course of the simulations. Conclusion: The results suggest that the proposed CNN-QSAR model can be an efficient method for hit prediction and a new way to identify hit compounds against 3CLpro of SARS-CoV-2.