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

Area of Science:

• Computational chemistry
• Molecular dynamics
• Free energy calculations

Background:

• Accurate free energy calculations require precise potential energy surfaces and efficient sampling.
• Neural network potentials (NNPs) offer higher accuracy than molecular mechanics (MM) force fields but are computationally expensive.
• Indirect free energy calculation (end-state correction) uses MM potentials for initial calculations and NNPs for reweighting.

Purpose of the Study:

• To systematically investigate the accuracy and robustness of reweighting from MM to NNP (ANI-2x) potentials.
• To assess the impact of simulation parameters on free energy calculations using reweighting techniques.
• To compare nonequilibrium (NEQ) switching simulations with equilibrium free energy simulations.

Main Methods:

• Single-step free-energy perturbation (FEP) and NEQ switching simulations were employed.
• Reweighting techniques were used to transfer results from MM to NNP potentials.
• The study analyzed the influence of switching lengths and slow degrees of freedom on work distribution outliers.

Main Results:

• NEQ switching simulations between MM and NNP potentials provide accurate free energy estimates.
• Longer switching lengths and slow degrees of freedom can impact outlier work distributions.
• NEQ switching simulations are more suitable than equilibrium methods for this MM-NNP transfer.

Conclusions:

• Nonequilibrium switching simulations are the preferred method for accurate free energy calculations involving NNPs and MM potentials.
• The NEQ switching approach is efficient, robust, and easy to implement for transferring calculations between MM and NNPs.