Using Neural Network Potentials to efficiently calculate indirect free energy estimates

Abstract

To achieve chemical accuracy in free energy calculations, it is necessary to accurately describe the system's potential energy surface and efficiently sample configurations from its Boltzmann distribution. While neural network potentials (NNPs) have shown significantly higher accuracy than classical molecular mechanics (MM) force fields, they have limited range of applicability and are significantly slower than MM potentials, often by orders of magnitude. To address this challenge, Rufa et al suggested a two-stage approach that uses a fast and established MM alchemical energy protocol, followed by reweighting the results using NNPs, known as endstate correction or indirect free energy calculation. In this study, we systematically investigate the accuracy, robustness, and efficiency of free energy calculations between a MM reference and a neural network target potential (ANI-2x) using free energy perturbation (FEP) and non-equilibrium (NEQ) switching simulation in vacuum. We investigate the influence of longer switching lengths and the impact of slow degrees of freedom on outliers in the work distribution. We compare the results to multistage equilibrium free energy calculations (MFES) for an established dataset. Our results demonstrate that free energy calculations between NNPs and MM potentials should not be performed using FEP but require NEQ switching simulations to obtain accurate free energy estimates. NEQ switching simulations between MM and NNPs are highly efficient, robust, and trivial to implement.

Insights and Challenges in Correcting Force Field Based Solvation Free Energies Using A Neural Network Potential

“…Indirect free energy calculations use a computationally cheaper description of the potential energy (e.g., a molecular mechanics force field) and calculate the free energy contribution needed for changing to a more expensive description of the potential energy (e.g., a QM potential). The free energy difference between the different levels of theory can be calculated reliably using nonequilibrium switching techniques [17,23,37,40,44]. NNPs offer a tempting tradeoff between accuracy and speed compared to molecular mechanics and QM methods, which is why they can be applied as the high-level potential in these indirect cycles.…”

“…An early example of using NNPs to refine classical free energy simulations is a study by Rufa et al [35]. Recently, we investigated the convergence of the correction step required in indirect pathways, i.e., calculating the free energy difference between an MM and a machine learning (ML) representation of a system [40]. In both studies, the ANI-2x NNP was used [10,39].…”

“…The computational framework of reference [40] is suitable for use in the gas phase and aqueous solution. In this work, we explore whether MM→ML / MM→ML/MM corrections can improve the agreement of computed solvation free energies with the experiment.…”