AI discovers new COVID drug candidates as variants threaten approved treatments
Researchers used machine learning and computational chemistry to identify six promising compounds that block a key coronavirus enzyme, with one showing potency comparable to existing drugs. The work signals a shift toward AI-accelerated drug discovery and addresses growing concerns about variant resistance to current therapies like nirmatrelvir.
Originaltitel: Structure- and Ligand-Based Discovery of Novel 3-Chymotrypsin-Like Protease Nonpeptidomimetic Hits.
The SARS-CoV-2 3-chymotrypsin-like (3CLpro) protease is a key target for the development of COVID-19 therapeutics. While ensitrelvir and nirmatrelvir are approved drugs for treatment, the continuous research and development for new antiviral drugs is necessary to combat the emergence of variants and other related viruses. This study employed structure- and ligand-based computer-assisted approaches to identify new 3CLpro nonpeptidomimetic inhibitors. Using data from COVID Moonshot, NCATS, and the literature, computational methods such as shape-based, ensemble docking, and machine learning (ML) techniques were developed, achieving robust validation metrics: AUC=87%, EF=7, BEDROC=60% for shape-based; AUC=87%, EF=7.03, BEDROC=62% for ensemble docking, and ACC=81%, MCC=62% for ML models, combing Random forest+ECFP4 fingerprint. These models were utilized in virtual screening (VS) campaigns using the H3D and ChemBridge libraries, from which six promising hits with IC50 values≤80µM were identified, including LabMol-499 with an IC50 of 13.71µM and a Ki of 21.74µM. Moreover, we found that LabMol-499 acts as a noncompetitive, reversible inhibitor of 3CLpro. These findings provide a foundation for hit-to-lead optimization of new nonpeptidomimetic 3CLpro inhibitors.