Quantum-based machine learning and AI models to generate force field parameters for drug-like small molecules

Area of Science:

• Computational chemistry
• Drug discovery
• Machine learning

Background:

• Force fields are crucial for molecular dynamics simulations and binding free energy calculations.
• Accurate partial charges on small molecules are essential for understanding protein-drug interactions.
• Generating these charges is currently a time-consuming process.

Purpose of the Study:

• To develop a machine learning (ML) model for rapid prediction of partial charges on small molecules.
• To create an AI-generated force field for drug-like molecules.
• To accelerate the generation of accurate force field parameters.

Main Methods:

• Performed density functional theory (DFT) calculations on 31,770 drug-like small molecules.
• Trained an ML model on DFT-based atomic charges to predict partial charges.
• Developed neural network (NN) models for atom types, phase angles, and periodicities.
• Calculated solvation free energies to assess force field accuracy.

Main Results:

• The ML model accurately predicted partial charges, comparable to DFT calculations.
• NN models demonstrated high accuracy for atom typing and other parameters.
• Calculated solvation free energies closely matched experimental values.
• The AI-generated force field enabled fast and accurate parameter generation.

Conclusions:

• The developed AI-generated force field significantly accelerates the prediction of partial charges and other parameters for small molecules.
• This approach enhances the efficiency of molecular dynamics simulations and binding free energy calculations in drug discovery.
• The ML and NN models provide a robust and accurate method for generating essential force field components.