Abstract:Machine Learning (ML) methods have reached high accuracy levels for the prediction of in vacuo molecular properties. However, the simulation of large systems through solely ML methods (like those based on neural network potentials) is still a challenge. In this context, one of the most promising frameworks for integrating ML schemes in the simulation of complex molecular systems are the so-called ML/MM methods. These multiscale approaches combine ML methods with classical forcefields (MM), in the same spirit as the succesful hybrid quantum mechanics-molecular mechanics methods (QM/MM). The key issue for such ML/MM methods is the adequate description of the coupling between the region of the system described by ML and the region described at the MM level. In the context of QM/MM schemes, the main ingredient of the interaction is electrostatic, and the state of the art is the so called electrostatic-embedding. In this study, we analyze the quality of simpler mechanical embedding-based approaches, specifically focusing on their application within a ML/MM framework utilizing atomic partial charges derived in vacuo. Taking as reference electrostatic embedding calculations performed at a QM(DFT)/MM level, we explore different atomic charges schemes, as well as a polarization correction computed using atomic polarizabilites. Our benchmark data set comprises a set of about 80k small organic structures from the ANI-1x database, solvated in water. The results suggest that the MBIS atomic charges yield the best agreement with the reference coupling energy. Remarkable enhancements are achieved by including a simple polarization correction.

What problem does this paper attempt to address?

This paper discusses how to effectively combine machine learning (ML) methods with classical force fields (MM) in molecular simulations to solve simulation problems of large and complex systems. Specifically, the focus of the research is on evaluating the performance of different embedding schemes in hybrid ML/MM frameworks, particularly Mechanical Embedding (ME) methods and Polarizable Mechanical Embedding (ME+P) methods. The paper compares and analyzes various atomic partial charge schemes and introduces polarization corrections to improve the accuracy of mechanical embedding. In traditional quantum mechanics/molecular mechanics (QM/MM) methods, electrostatic embedding is the most commonly used coupling method, which uses QM electronic density to calculate interactions between QM and MM regions. However, mechanical embedding does not consider the influence of MM systems on QM energy and forces; instead, it preassigns atomic charges to estimate coupling energies. In order to approach the quality of electrostatic embedding, some methods attempt to include MM systems within the ML framework or use atomic polarizability to correct energies. In this paper, the researchers used a database of around 80,000 small organic structures dissolved in water as a benchmark test set, with DFT electrostatic embedding calculations as a reference, to study the effects of different atomic charge schemes and polarization corrections. The results demonstrated that the MBIS atomic charge scheme was most consistent with the reference coupling energy, and the accuracy of the results could be significantly improved by a simple polarization correction. In conclusion, the paper aims to improve ML/MM methods, particularly by optimizing mechanical embedding schemes and introducing polarization effects, to more accurately simulate large and complex molecular systems, such as solvent effects and molecular interactions in enzymatic catalysis and other biochemical processes.

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

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

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

Advancing Multiscale Molecular Modeling with Machine Learning-Derived Electrostatics

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

An evaluation of machine learning/molecular mechanics end-state corrections with mechanical embedding to calculate relative protein-ligand binding free energies

Learning QM/MM Potential using Equivariant Multiscale Model

Learning the Quantum Centroid Force Correction in Molecular Systems: A Localized Approach

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.

A quantum mechanical computational method for modeling electrostatic and solvation effects of protein

Accurate Free Energy Calculation via Multiscale Simulations Driven by Hybrid Machine Learning and Molecular Mechanics Potentials

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

A Hessian-Based Assessment of Atomic Forces for Training Machine Learning Interatomic Potentials

Asymptotic Expansion for Electrostatic Embedding Integrals in QM/MM Calculations.

Analytic gradients for the electrostatic embedding QM/MM in periodic boundary conditions using particle-mesh Ewald sums and electrostatic potential fitted charge operators

Density Embedded VB/MM: A Hybrid Ab Initio VB/MM with Electrostatic Embedding

Accurate interatomic force field for molecular dynamics simulation by hybridizing classical and machine learning potentials

Assessment of free energies from electrostatic embedding density functional tight binding-based/molecular mechanics in periodic boundary conditions

Multistate, polarizable QM/MM embedding scheme based on the direct reaction field method: Solvatochromic shifts, analytical gradients and optimization of conical intersections in solution

Challenges for Machine Learning Force Fields in Reproducing Potential Energy Surfaces of Flexible Molecules