Machine learning based implicit solvent model for aqueous-solution alanine dipeptide molecular dynamics simulations.

Area of Science:

• Computational Chemistry
• Molecular Dynamics
• Machine Learning

Background:

• Implicit solvent models simplify molecular simulations by approximating solvent effects.
• Developing accurate implicit solvent models from molecular dynamics (MD) data is challenging.

Purpose of the Study:

• To investigate the use of machine learning (ML) techniques to derive an implicit solvent model directly from explicit solvent MD simulations.
• To develop accurate ML-based implicit solvent models for QM/MM and ab initio-QM MD simulations.

Main Methods:

• Trained a machine learning potential (MLP) using the DeepPot-SE representation for alanine dipeptide.
• Captured solute-solvent interactions using the average solvent environment configuration (ASEC).
• Validated MLP accuracy against explicit solvent MD simulations and QM/MM forces.

Main Results:

• MLP predicted forces deviated by only 0.4 kcal mol⁻¹ Å⁻¹ from reference values.
• MLP-based free energy surface differed by less than 0.9 kcal mol⁻¹ from explicit MD results.
• MLP accurately reproduced QM/MM forces in an ASEC environment.

Conclusions:

• ML techniques can successfully derive accurate implicit solvent models from explicit solvent MD simulations.
• The developed ML-based implicit solvent models are cost-effective for training and inference.
• This method enables accurate and efficient ML-based implicit solvent models for QM calculations.