Computing Solvation Free Energies of Small Molecules with Experimental Accuracy

Area of Science:

• Computational chemistry
• Molecular dynamics simulations
• Machine learning in chemistry

Background:

• Free energies are crucial for understanding chemical systems and are key outputs of molecular dynamics simulations.
• Solvation free energies of drug-like molecules are vital for assessing force field accuracy and predicting protein-ligand binding.
• Machine learned potentials (MLPs) offer higher accuracy than empirical force fields but are difficult to integrate with standard alchemical free energy methods.

Purpose of the Study:

• To develop an efficient alchemical free energy protocol compatible with machine learned potentials (MLPs).
• To enable rigorous free energy difference calculations in condensed phase systems modeled entirely by MLPs.
• To demonstrate the potential of MLPs in overcoming limitations of empirical force fields for free energy calculations.

Main Methods:

• Introduction of an efficient alchemical free energy protocol for systems modeled by MLPs.
• Utilizing a pretrained, transferable, and alchemically equipped MLP model.
• Performing condensed phase free energy calculations entirely within an MLP framework.

Main Results:

• Successful implementation of a rigorous alchemical free energy protocol using MLPs.
• Demonstration of subchemical accuracy in solvation free energy calculations.
• Application to a diverse set of organic molecules.

Conclusions:

• Machine learned potentials can be effectively integrated into alchemical free energy calculations.
• The developed protocol enables accurate prediction of solvation free energies, addressing limitations of empirical force fields.
• This work paves the way for more accurate computational chemistry simulations using MLPs.