Solvent-Inclusive ML/MM Simulations: Assessments of Structural, Dynamical, and Thermodynamic Accuracy.

Area of Science:

• Computational chemistry
• Chemical physics
• Biophysics

Background:

• Accurate modeling of chemical reactions in solution is crucial for various scientific fields.
• Hybrid machine-learned/molecular mechanics (ML/MM) simulations offer a balance between accuracy and efficiency.
• Existing ML/MM methods often exclude solvent molecules, limiting their ability to capture solvent-mediated reactivity.

Purpose of the Study:

• To develop and evaluate a novel solvent-inclusive ML/MM methodology.
• To address the limitations of current ML/MM approaches in modeling solvent effects.
• To enable more accurate simulations of chemical reactions influenced by solvent.

Main Methods:

• Introduced a solvent-inclusive ML/MM framework with fixed spatial boundaries for ML and MM regions.
• Employed a force-partitioning scheme to manage interactions across the ML/MM boundary.
• Evaluated the methodology through simulations of bulk water and formic acid dissociation.

Main Results:

• Structural and dynamical properties of bulk water were preserved with sufficiently large ML regions.
• The choice of potential energy surface representation impacted properties, especially at smaller ML region sizes.
• Free energy profiles for formic acid dissociation showed trends consistent with reference systems, with some observed deviations.

Conclusions:

• The developed solvent-inclusive ML/MM approach shows promise for modeling solvent-mediated reactivity.
• Sufficiently large ML regions are necessary to maintain accurate water properties.
• Further development is needed to address observed deviations and refine the methodology.