Solvation Free Energies from Neural Thermodynamic Integration (2410.15815v3)

Abstract: We present a method for computing free-energy differences using thermodynamic integration with a neural network potential that interpolates between two target Hamiltonians. The interpolation is defined at the sample distribution level, and the neural network potential is optimized to match the corresponding equilibrium potential at every intermediate time-step. Once the interpolating potentials and samples are well-aligned, the free-energy difference can be estimated using (neural) thermodynamic integration. To target molecular systems, we simultaneously couple Lennard-Jones and electrostatic interactions and model the rigid-body rotation of molecules. We report accurate results for several benchmark systems: a Lennard-Jones particle in a Lennard-Jones fluid, as well as the insertion of both water and methane solutes in a water solvent at atomistic resolution using a simple three-body neural-network potential.

• The paper introduces a neural-network driven thermodynamic integration technique that accurately estimates solvation free energies by training on potential energy interpolants.
• The method leverages stochastic interpolants to bypass extensive sampling, yielding results that align with traditional thermodynamic integration outcomes.
• The research implies that integrating AI with classical methods can reduce computational costs and boost simulation throughput for complex molecular systems.

Overview of "Solvation Free Energies from Neural Thermodynamic Integration"

This paper introduces an innovative approach to calculate solvation free energies by utilizing neural thermodynamic integration (NTI), incorporating a neural-network potential that interpolates between two Hamiltonians. Addressing a central computational challenge in physics, chemistry, and biology, this approach offers crucial insights into processes that govern molecular stability and reactions.

Methodological Advancement

The authors propose a novel methodology that extends traditional thermodynamic integration (TI) by introducing stochastic interpolants into the framework. This approach enhances flexibility compared to conventional methods which rely heavily on Monte Carlo or molecular dynamics simulations across a predefined trajectory. The key lies in training a neural network to approximate the potential energy function at different stages of interpolation, thus facilitating the estimation of free energy differences between two states.

Validation and Performance

The paper validates the proposed NTI technique on benchmark systems, including solvation phenomena involving Lennard-Jones particles and atomistically resolved water and methane solutes in water solvent. Results indicate that the computed solvation free energies show excellent congruence with reference values, confirming the efficacy of NTI. Notably, the free energy differences calculated by the neural network align well with those obtained from traditional TI, despite NTI's computational efficiency by eliminating the need for intermediate sampling across the interpolation path.

Implications and Speculation on Future Directions

The implications of this research are significant for computational modeling in molecular sciences. By leveraging machine learning, responses from complex molecular systems can be obtained with reduced computational cost, enhancing simulation throughput and scalability. The flexibility and potential extension of NTI to more complicated reactive systems suggest broad applicability, opening avenues for investigating free energy landscapes in diverse material and biological systems.

The work potentially marks a shift towards integrating AI with classical thermodynamic methods. Future developments may see NTI being expanded to accommodate non-rigid molecular systems or even more complex solute-solvent interactions. As AI continues to mature, its synergy with simulation techniques like NTI could lead to more precise and comprehensive modeling paradigms for free energy calculations in complex systems. This intersection of AI and thermodynamics holds promise for advancing both theoretical frameworks and practical computational tools in molecular and material sciences.