Multimolecular feature-based machine learning: system biology enhanced RF-QSAR modeling for the efficient prediction of the inhibitory potential of diverse SARS CoV-2 3CL Protease inhibitors

The COVID-19 pandemic has catalyzed an urgent need for effective treatment, given ongoing concerns about vaccine efficacy. By leveraging the CHEMBL database, we focused on SARS-CoV-2 3C-like protease (3CLpro) inhibitors, which are crucial drug targets in the coronavirus genome. We deployed a multi-faceted strategy combining quantitative structure-activity relationship (QSAR) analysis, molecular docking, dynamic simulations, and system biology to discover promising antiviral compounds. This culminated in a cheminformatics pipeline that crafted QSAR models, powered by machine learning. These models analyzed 919 molecules and demonstrated outstanding correlation coefficients of 97.36% and 74.13%. The models were built on the foundation of substructure fingerprints and 1D/2D molecular descriptors. Tools such as variance importance plots (VIPs) and correlation matrices were utilized to identify key features crucial for inhibiting 3CL protease. A web tool, 3CLpro-pred -Selec-Pred ( https://3clpro-pred.streamlit.app/ ), was developed for streamlined bioactivity prediction against the protease. Molecular docking and dynamics provided atomic-level insights, confirming QSAR hypotheses and improving understanding of interaction dynamics. Hits compounds from QSAR and molecular modeling studies were further evaluated by system biology to elucidate the mechanisms underlying potential compounds’ inhibitory effects on SARS-CoV-2 3CL protease. Gene ontology and KEGG pathway analysis identified the most relevant pathways for the involved genes. Detailed system biology analysis highlighted crucial genes like TBK1, PIK3CA, IKBKB, GSK3B, and CASP3 as key nodes, linking compounds, targets, and pathways. The identified molecular interactions invite more extensive experimental verification to validate their therapeutic promise in antiviral drug discovery. Highlights Built ML-QSAR models with fingerprints and descriptors, effectively pinpointing potential inhibitors Made a user-friendly web tool, “3CLpro-pred -Selec-Pred” ( https://3clpro-pred.streamlit.app/ ) to quickly identify potential drugs from over 1200 tested, finding 82 promising ones. Developed an advanced RF-QSAR model enhanced by system biology to accurately predict potential COVID-19 treatments targeting SARS-CoV-2 3CL Protease. Compound F0648-0705 showed good binding in the active site of the SARS-COV-2 3CL protease. Showed how studying the complex interactions between drugs, genes, and the virus’s life cycle can reveal new ways to fight the virus, emphasizing the need for further testing to confirm these findings.