Enhancing Protein-Ligand Binding Affinity Predictions using Neural Network Potentials.

Area of Science:

• Computational chemistry and molecular modeling.
• Drug discovery and development.
• Biophysics and structural biology.

Background:

• Accurate prediction of protein-ligand binding affinity is crucial for drug discovery.
• Conventional molecular mechanics (MM) force fields often struggle with precision in these predictions.
• Molecular dynamics (MD) simulations offer a powerful tool but require accurate energy potentials.

Approach:

• Developed a hybrid neural network potential and molecular mechanics (NNP/MM) methodology.
• Employed machine learning potentials to enhance the accuracy of MD simulations.
• Utilized the Alchemical Transfer Method (ATM) for computing relative binding free energies (RBFE).

Key Points:

• The NNP/MM approach demonstrates significant improvements in predicting protein-ligand binding affinities.
• Relative binding free energy (RBFE) calculations using ATM showed enhanced performance.
• Validated against established benchmarks, confirming the methodology's robustness.

Conclusions:

• The hybrid NNP/MM method offers a substantial advancement over traditional MM force fields like GAFF2 for binding affinity predictions.
• This approach holds promise for accelerating the identification of potent drug candidates.
• Machine learning potentials are increasingly vital for high-accuracy molecular simulations.