Automated Refinement of Property-Specific Polarizable Gaussian Multipole Water Models Using Bayesian Black-Box...

Area of Science:

• Computational Chemistry
• Biomolecular Simulations
• Physical Chemistry

Background:

• Accurate water models are essential for simulating biochemical processes.
• The polarizable Gaussian multipole (pGM) model offers improved handling of biomolecular interactions.

Purpose of the Study:

• To develop and optimize a three-center pGM water model for enhanced biomolecular simulations.
• To refine van der Waals and electrostatic parameters for accurate reproduction of liquid water properties.

Main Methods:

• Utilized ab initio quantum mechanical calculations for initial model exploration.
• Employed automated machine learning (AutoML) for optimizing the pGM model parameters.
• Validated the model using liquid-phase water properties at 298 K and 1.0 bar.

Main Results:

• Developed the pGM3P-25 model, demonstrating marked enhancements in accuracy and practical utility.
• Accurately reproduced key properties like oxygen-oxygen radial distribution function, density, and dipole moment.
• Successfully predicted thermodynamic and temperature-dependent properties not explicitly used in training.

Conclusions:

• The pGM3P-25 model shows significant improvements for molecular dynamics simulations.
• AutoML frameworks streamline parameter refinement, reducing time and human effort.
• This approach broadens the applicability of molecular dynamics in computational chemistry.