AI-Augmented Virtual Screening for COVID-19 Antivirals

Abstract

The ongoing global threat of COVID-19 continues to require speedy and dependable antiviral drug discovery techniques. Traditional drug-screening protocols are time-consuming and very much limited in predictive shortcomings, while AI-driven methods, although computationally intensive, can be lacking in molecular binding's structural elements. This research introduces an AI-facilitated virtual screening strategy marrying machine learning–based affinity prediction with structure-based molecular docking for identifying potent inhibitors of the SARS-CoV-2 main protease (Mpro). A curated dataset of 4,200 drug-like compounds, sourced from ChEMBL and PubChem and filtered using Lipinski’s Rule of Five, was utilized for model training and validation. A Random Forest ensemble that was trained using molecular descriptors had a prediction accuracy of 91.3%, outperforming both traditional docking (68.4%) and AI-exclusive models (82.7%) in prediction precision, valid binding hits, and screening efficiency. The double-score integration and double-validation strategy successfully minimized false positives by synergizing statistical learning and structural validation. The hybrid approach improves computational efficiency and interpretability, with a reproducible and extensible methodology towards antiviral discovery. Future developments involve integrating dynamic protein modeling, ADMET profiling, and improved compound diversity to further enhance biological relevance and predictive robustness.

Ähnliche Arbeiten