Multiscale Force Field Model Based on a Graph Neural Network for Complex Chemical Systems.

Area of Science:

• Computational Chemistry
• Multiscale Modeling
• Machine Learning in Science

Background:

• The QM/MM methodology offers a balance between accuracy and efficiency in simulations.
• Accelerating quantum mechanical (QM) calculations using machine learning (ML) potential energy surfaces is a key advancement.
• Designing effective interactions between ML and molecular mechanics (MM) regions remains a challenge.

Purpose of the Study:

• To develop a novel ML/MM approach for multiscale simulations.
• To accurately model electrostatic interactions between ML and MM regions.
• To enhance the computational efficiency of QM/MM methods.

Main Methods:

• Utilized a graphical neural network based on stationary perturbation theory for electrostatic interactions.
• Processed atomic coordinates and MM charges to compute electrostatic energy and forces.
• Developed a solvent-free protocol for dataset preparation.

Main Results:

• Achieved a high-performance electrostatic embedding ML/MM architecture.
• Validated the accuracy of ML/MM energy calculations in aqueous solutions.
• Demonstrated parameter transferability across diverse solvent environments, including ionic liquids and interfaces.
• Observed the catalytic effect of aqueous solutions on the Claisen rearrangement of AVE.

Conclusions:

• The developed ML/MM approach effectively models electrostatic interactions, enhancing computational efficiency.
• The method shows promising transferability across various chemical environments.
• This work provides a robust framework for advanced multiscale simulations in chemistry and materials science.