Reweighting from Molecular Mechanics Force Fields to the ANI-2x Neural Network Potential

• DOI: 10.1021/acs.jctc.3c01274
• Publication Date: 2024-03-26T03:34:26Z
• Authors: Sara Tkaczyk; Johannes Karwouno...; Andreas Schöller; H. Lee Woodcock; Thierry Langer; Stefan Boresch; Marcus Wieder
• Keywords: Robustness; Free energy perturbation

To achieve chemical accuracy in free energy calculations, it is necessary to accurately describe the system's potential surface and efficiently sample configurations from its Boltzmann distribution. While neural network potentials (NNPs) have shown significantly higher than classical molecular mechanics (MM) force fields, they a limited range of applicability are considerably slower MM potentials, often by orders magnitude. address this challenge, Rufa et al. [Rufa bioRxiv 2020, 10.1101/2020.07.29.227959.] suggested two-stage approach that uses fast established alchemical protocol, followed reweighting results using NNPs, known as endstate correction or indirect calculation. This study systematically investigates robustness an reference target (ANI-2x) for data set vacuum, single-step free-energy perturbation (FEP) nonequilibrium (NEQ) switching simulation. We assess influence longer lengths impact slow degrees freedom on outliers work distribution compare those multistate equilibrium simulations. Our demonstrate calculations between NNPs should be preferably performed NEQ simulations obtain accurate estimates. efficient, robust, trivial implement.