QSAR‐Based Machine Learning Reveals a Repurposed Dual‐Function G4 Ligand against Ensitrelvir‐Resistant SARS‐CoV‐2 Main Protease

Abstract The emergence of drug‐resistant SARS‐CoV‐2 variants necessitates novel antiviral strategies targeting conserved viral components. This study integrates machine learning‐based quantitative structure‐activity relationship modeling and comprehensive computational approaches to identify dual‐function inhibitors against the main protease and RNA G‐quadruplex structures of SARS‐CoV‐2. A Random Forest classifier trained on 890 curated compounds achieves superior predictive performance (AUC = 0.9458) using CDK fingerprints, enabling virtual screening of 4,564 G‐quadruplex ligands from the G4LDB. Molecular docking reveals lead compound G4L2574 exhibits stronger binding affinity (−12.11 kcal mol −1 ) to the M49I mutant Mpro than clinical inhibitor ensitrelvir (−8.92 kcal mol −1 ), with molecular dynamics simulations demonstrating enhanced complex stability and persistent hydrogen bonding. MM/PBSA calculations confirm favorable binding free energy (−40.54 kcal mol −1 ) for G4L2574‐M49I, driven by robust electrostatic interactions. Structural analysis shows the M49I mutation induced steric hindrance compromising ensitrelvir binding, while G4L2574 maintained critical interactions with catalytic residues His41 and Cys145. Additionally, G4L2574 demonstrates superior RNA G‐quadruplex binding (−11.73 kcal mol −1 ) than RNA G‐quadruplex stabilizing ligand TMPyP4. This dual‐targeting mechanism, validated through machine learning and MD simulations, presents a promising strategy to circumvent resistance mutations while leveraging conserved viral replication targets. The integrated computational pipeline establishes a framework for rapid identification of broad‐spectrum antivirals against evolving coronaviruses.