Unveiling the antiviral inhibitory activity of ebselen and ebsulfur derivatives on SARS-CoV-2 using machine learning-based QSAR, LB-PaCS-MD, and experimental assay

Ebsulfur and ebselen derivatives that were proven to be potent inhibitors against the main protease (M^Pro) of SARS-CoV-2 which is an essential enzyme for viral replication were chosen to study the quantitative structure-activity relationship (QSAR) analysis using a classical multiple linear regression (MLR) and a machine learning approach of random forest (RF) and artificial neural network (ANN) in order to find the relationship between molecular structural properties and biological inhibitory activities. With the statistical criteria, the R² values of MLR, RF, and ANN models for the training set were 0.83, 0.82, and 0.92, respectively. The RMSE values of the test were considered for model evaluation, and the results were 0.27, 0.18, and 0.09 for MLR, RF, and ANN models, respectively. Therefore, the ANN model was the best-obtained model for predicting the M^Pro inhibitory activity of thirteen new synthetic ebselen analogs that haven't tested the biological assay before. Notably, our predicted inhibitory activities against SARS-CoV-2 were then examined using enzyme-based assays and cytotoxicity tests, which found that compound P8 resulted in a good potential candidate for SARS-CoV-2 M^Pro inhibitory activity. Furthermore, the molecular dynamics simulations were performed to study the dynamic interaction of ligand and binding site; the results showed a binding pathway and mechanism of compound P8 with key residues surrounding the active site of SARS-CoV-2 M^Pro, which is useful for further development of ebselen derivatives.