Quantum Machine Learning Corrects Classical Force Fields: Stretching DNA Base Pairs in Explicit Solvent

TL;DR

This paper introduces a quantum machine learning approach to enhance classical force fields in molecular dynamics, specifically improving DNA stretching simulations by accounting for quantum effects and explicit solvation.

Contribution

The study presents a novel kernel-based quantum correction method integrated into molecular dynamics, improving the accuracy of DNA stretching models beyond classical force fields.

Findings

Classical DNA models are overly stiff in stretching simulations.

Quantum corrections align better with experimental thermodynamics.

The KMMD method is implemented in AMBER22 software.

Abstract

In order to improve the accuracy of molecular dynamics simulations, classical force fields are supplemented with a kernel-based machine learning method trained on quantum-mechanical fragment energies. As an example application, a potential-energy surface is generalised for a small DNA duplex, taking into account explicit solvation and long-range electron exchange--correlation effects. Study of the corrected potential energy versus extension shows that leading classical DNA models have excessive stiffness with respect to stretching. This discrepancy is found to be common across multiple forcefields. The quantum correction is in qualitative agreement to the experimental thermodynamics for larger DNA double helices, providing a candidate explanation for the general and long-standing discrepancy between single molecule stretching experiments and classical calculations of DNA stretching. The new dataset of quantum calculations and the associated Kernel Modified Molecular Dynamics (KMMD) method should be of general utility in biomolecular simulations. KMMD is made available as part of the AMBER22 simulation software.