AI-Assisted Virtual Screening and ADMET-Guided Prioritisation of South Indian Plant-Derived Phytochemicals Against the SARS-CoV-2 Main Protease (Mpro): A Reproducible Computational Framework and Curated Compound Library

Abstract

Background: The main protease (M^pro; 3CL^pro; nsp5) of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) is an essential, highly conserved enzyme that governs viral polyprotein maturation and has no close human homologue, making it a premier antiviral target. Plant-derived phytochemicals, many rooted in the ethnomedicine of South India, provide a chemically diverse starting point for inhibitor discovery. We assembled a curated library and established a transparent, reproducible artificial-intelligence (AI)-assisted virtual-screening and ADMET-profiling framework centred on this target. Methods: Fifty phytochemicals with verified PubChem identities and representative South Indian botanical sources were compiled against the M^pro crystal structure 6LU7 (catalytic dyad His41/Cys145). Two-dimensional structures were curated as canonical SMILES, and physicochemical descriptors (molecular weight, calculated LogP, hydrogen-bond donors/acceptors, topological polar surface area [TPSA], rotatable bonds) were computed with RDKit. A tiered prioritisation cascade applied Lipinski and Veber drug-likeness rules and ranked survivors by the quantitative estimate of drug-likeness (QED). The docking, AI activity-probability, and predictive ADMET modules of the framework, together with the editable composite-score weighting scheme, are fully specified for downstream execution. Results: The library spanned 32 phytochemical classes, dominated by flavonols, anthraquinones, flavones and triterpenoids. Forty-nine compounds were descriptor-profiled (mean molecular weight 313 g/mol; mean cLogP 2.32); 45 satisfied the Lipinski rule of five (<=1 violation) and 40 additionally passed the Veber criteria. The four rejected scaffolds were large, highly hydroxylated glycosides and a triterpenoid saponin (rutin, hesperidin, epigallocatechin gallate, glycyrrhizin). The top QED-ranked candidates included zingerone, hesperetin, naringenin, ferulic acid, carnosic acid, daidzein, eugenol, resveratrol and berberine (QED 0.63-0.80). Conclusion: We present a verified 50-compound South Indian phytochemical library and a transparent, integrity-preserving computational pipeline for M^pro inhibitor discovery. Small phenolics, flavanones and isoflavones emerge as favourable, drug-like starting points warranting subsequent molecular docking, AI scoring, ADMET prediction and experimental validation.