Severe acute respiratory syndrome coronavirus-2 (SARS-CoV-2) is a member of the Coronaviridae family, causing major destructions to human life directly and indirectly to the economic crisis around the world. Although there is significant reporting on the whole genome sequences and updated data for the different receptors are widely analyzed and screened to find a proper medication. Only a few bioassay experiments were completed against SARS-CoV-2 spike protein. We collected the compounds dataset from the PubChem Bioassay database having 1786 compounds and split it into the ratio of 80-20% for model training and testing purposes, respectively. Initially, we have created 11 models and validated them using a fivefold validation strategy. The hybrid consensus model shows a predictive accuracy of 95.5% for training and 94% for the test dataset. The model was applied to screen a virtual chemical library of Natural products of 2598 compounds. Our consensus model has successfully identified 75 compounds with an accuracy range of 70-100% as active compounds against SARS-CoV-2 RBD protein. The output of ML data (75 compounds) was taken for the molecular docking and dynamics simulation studies. In the complete analysis, the Epirubicin and Daunorubicin have shown the docking score of -9.937 and -9.812, respectively, and performed well in the molecular dynamics simulation studies. Also, Pirarubicin, an analogue of anthracycline, has widely been used due to its lower cardiotoxicity. It shows the docking score of -9.658, which also performed well during the complete analysis. Hence, after the following comprehensive pipeline-based study, these drugs can be further tested in vivo for further human utilization.Communicated by Ramaswamy H. Sarma.
Keywords: SARS-CoV-2; deep screening; machine learning models; molecular dynamics simulation; receptor-binding domain.