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

Area of Science:

• Computational chemistry
• Molecular modeling
• Drug discovery

Background:

• Accurate prediction of protein-ligand binding affinity is crucial for drug discovery.
• Traditional molecular mechanics (MM) force fields often struggle with accuracy in binding affinity calculations.
• Molecular dynamics (MD) simulations offer a powerful framework for studying molecular interactions.

Purpose of the Study:

• To improve the accuracy of protein-ligand binding affinity predictions.
• To introduce and evaluate a novel hybrid machine learning potential and molecular mechanics (NNP/MM) methodology.
• To assess the performance of the NNP/MM approach against established benchmarks.

Main Methods:

• Utilizing molecular dynamics (MD) simulations.
• Employing a hybrid neural network potential and molecular mechanics (NNP/MM) methodology.
• Calculating relative binding free energies using the Alchemical Transfer Method.

Main Results:

• The NNP/MM methodology demonstrated significant enhancements in predicting protein-ligand binding affinities.
• The developed approach outperformed conventional MM force fields, such as GAFF2.
• Validation against established benchmarks confirmed the reliability and improved performance of the NNP/MM method.

Conclusions:

• The hybrid NNP/MM approach represents a substantial advancement in computational drug discovery.
• This method offers a more accurate and reliable way to predict protein-ligand binding affinities.
• The findings pave the way for more efficient and effective lead optimization in pharmaceutical research.