Benchmarking ANI potentials as a rescoring function and screening FDA drugs for SARS-CoV-2 M^pro

Area of Science:

• Computational chemistry
• Drug discovery
• Molecular modeling

Background:

• Molecular docking is crucial for identifying drug candidates.
• Existing scoring functions have limitations in accuracy and computational efficiency.
• Accurate prediction of host-guest interactions is essential for reliable docking.

Purpose of the Study:

• To introduce and evaluate Artificial Intelligence Machine Learning (ANI-ML) potentials as a novel rescoring function in molecular docking.
• To compare the performance of ANI-ML potentials against established scoring functions.
• To demonstrate the utility of ANI-ML potentials in a drug screening campaign.

Main Methods:

• Utilized ANI-ML potentials as a rescoring function in molecular docking.
• Benchmarked ANI-ML potentials on the CASF-2016 dataset against 34 other scoring functions.
• Combined GOLD-PLP docking, ANI-ML rescoring, and molecular dynamics (MD) simulations with free energy methods for drug screening.
• Screened FDA-approved drugs against SARS-CoV-2 main protease (Mpro).

Main Results:

• ANI-ML potentials demonstrated competitive "docking power" with current scoring functions at similar computational costs.
• ANI-ML ranked in the top 5 scoring functions out of 34 tested on the CASF-2016 dataset.
• Combining ANI-ML with GOLD-PLP improved the accuracy of top-ranked docking solutions.
• The screening protocol identified six promising drug molecules against SARS-CoV-2 Mpro, consistent with previous studies.

Conclusions:

• ANI-ML potentials represent a significant advancement in molecular docking scoring functions.
• The accuracy and efficiency of ANI-ML facilitate large-scale drug candidate screening.
• The validated screening methodology holds promise for identifying novel therapeutics for diseases like COVID-19.