Integrated machine learning-driven QSAR modelling and systems biology approach for the identification of potential SARS-CoV-2 3CLpro inhibitors

Abstract

SARS-CoV-2 3C-like protease (3CLpro) is an essential viral enzyme responsible for processing viral polyproteins and is a validated target for small-molecule therapeutic intervention. This study presents an integrative cheminformatics and systems biology framework to identify and evaluate potential 3CLpro inhibitors. A curated dataset of 919 compounds from the CHEMBL database was used to develop predictive QSAR models based on substructure fingerprints and 1D/2D molecular descriptors. The best-performing models achieved strong correlation coefficients (0.9736 for training and 0.7413 for testing), and key molecular features were identified using feature importance analysis. A web-based tool, 3CLpro-Pred was developed to enable rapid prediction of compound bioactivity. Molecular docking and dynamics simulations further validated QSAR findings by elucidating key atomic-level interactions at the protease active site. Top hit compounds were prioritized for systems-level analysis at the host-pathogen interface, where gene ontology and KEGG pathway enrichment revealed their potential to modulate key host signalling pathways. Critical regulatory genes, including TBK1, PIK3CA, IKBKB, GSK3B, and CASP3, were identified as major nodes linking compound action to antiviral and immune processes. This study delivers a combinatorial computational pipeline to accelerate the discovery of 3CLpro-targeted antivirals, providing mechanistic insights and a shortlist of candidates for future experimental validation.