Software Infrastructure for Next-Generation QM/MM-ΔMLP Force Fields

Area of Science:

• Computational Chemistry
• Molecular Dynamics
• Machine Learning

Background:

• Accurate molecular simulations require robust force fields.
• Integrating quantum mechanics and machine learning offers enhanced accuracy.
• Existing methods face limitations in speed and applicability.

Purpose of the Study:

• To present novel software infrastructure for designing and testing QM/MM-ΔMLP force fields.
• To validate the implementation in molecular dynamics and free energy simulations.
• To enable the study of complex biomolecular systems.

Main Methods:

• Integration of Amber, xtb (GFN2-xTB), and DeePMD-kit.
• Development of interfaces for QM/MM simulations with machine-learning corrections.
• Application in molecular dynamics and free energy calculations.

Main Results:

• Successful implementation of QM/MM-ΔMLP force field design and testing infrastructure.
• Demonstration of utility through proof-of-concept applications.
• Validation in molecular dynamics and free energy simulations.

Conclusions:

• The open-source software provides a powerful platform for developing new QM/MM-ΔMLP models.
• This technology facilitates research in biomolecular reactivity and protein-ligand binding.
• The infrastructure is freely available for broad scientific application.