Abstract:We present in this work the emle-engine package (https://github.com/chemle/emle-engine)─the implementation of a new machine learning embedding scheme for hybrid machine learning potential/molecular-mechanics (ML/MM) dynamics simulations. The package is based on an embedding scheme that uses a physics-based model of the electronic density and induction with a handful of tunable parameters derived from in vacuo properties of the subsystem to be embedded. This scheme is completely independent of...

What problem does this paper attempt to address?

This paper introduces a software package called emle-engine, which implements a new machine learning embedding scheme for hybrid machine learning potential/molecular mechanics simulations. The scheme is based on physically motivated electron density and induction models, requiring only a few adjustable parameters that come from the vacuum properties of the subsystem. This method is independent of the vacuum potential and only requires the atom positions in the machine learning subsystem and the charge positions in the molecular mechanics environment. By demonstrating its stability in enhanced sampling molecular dynamics simulations, the paper proves that the implemented electrostatic machine learning embedding (referred to as EMLE) surpasses traditional charge-fixed-based molecular mechanics embedding. The paper validates the performance of EMLE by comparing the free energy surfaces of alanine dipeptide in water under different embedding models. Compared to the reference DFT/MM free energy surface, EMLE's embedding method significantly reduces the average error because it takes into account the configuration dependence and the induced energy of the electron density. The paper also discusses two main advantages of EMLE: first, it can utilize pre-trained machine learning potentials without further training; second, it only requires information provided by existing QM/MM software, namely atom positions and charges, enabling seamless integration with existing software. Furthermore, the paper presents the theoretical background of the EMLE model, including how to predict total energy, static and induced potentials, and implementation details. By integrating with the sander program, this framework is applied to compute the alanine dipeptide free energy surface and compared with other models. Finally, the researchers discuss the results and outlook future work, emphasizing the potential of EMLE in handling systems and processes with significant charge distribution changes as well as its practical value in ML/MM simulations.

emle-engine : A Flexible Electrostatic Machine Learning Embedding Package for Multiscale Molecular Dynamics Simulations

emle-engine: a flexible electrostatic machine learning embedding package for multiscale molecular dynamics simulations

Electrostatic Embedding Machine Learning for Ground and Excited State Molecular Dynamics of Solvated Molecules

Assessment of Embedding Schemes in a Hybrid Machine Learning/Classical Potentials (ML/MM) Approach

Assessment of embedding schemes in a hybrid machine learning/classical potentials (ML/MM) scheme

Machine learning force fields for molecular liquids: Ethylene Carbonate/Ethyl Methyl Carbonate binary solvent

Synergistic integration of physical embedding and machine learning enabling precise and reliable force field

MACE: A Machine learning Approach to Chemistry Emulation

ANI Neural Networks Meet Electrostatics: A ML/MM Implementation in Amber

LVC/MM: A Hybrid Linear Vibronic Coupling/Molecular Mechanics Model with Distributed Multipole-Based Electrostatic Embedding for Highly Efficient Surface Hopping Dynamics in Solution

Machine learning for molecular simulation

Machine Learning in QM/MM Molecular Dynamics Simulations of Condensed-Phase Systems

Accelerated Quantum Mechanics/Molecular Mechanics Simulations Via Neural Networks Incorporated with Mechanical Embedding Scheme.

Molecular Simulations with a Pretrained Neural Network and Universal Pairwise Force Fields

Molecular Dynamics with On-the-Fly Machine Learning of Quantum-Mechanical Forces

Efficient Machine Learning Force Field for Large-Scale Molecular Simulations of Organic Systems

An Atomistic Machine Learning Package for Surface Science and Catalysis

sGDML: Constructing Accurate and Data Efficient Molecular Force Fields Using Machine Learning

mechanoChemML: A software library for machine learning in computational materials physics

Biomolecular dynamics with machine-learned quantum-mechanical force fields trained on diverse chemical fragments

A deep learning framework to emulate density functional theory