PhysNet meets CHARMM: A framework for routine machine learning/molecular mechanics simulations

Area of Science:

• Computational Chemistry
• Molecular Dynamics
• Spectroscopy

Background:

• Machine learning (ML) techniques are increasingly used to generate accurate potential energy surfaces (PESs) for molecular simulations.
• Integrating ML-based PESs into established simulation packages like pyCHARMM is crucial for practical applications.
• Accurate simulations are vital for understanding molecular behavior in various phases and predicting experimental observables.

Purpose of the Study:

• Introduce the MLpot extension with PhysNet for generating ML-based PESs within the pyCHARMM framework.
• Demonstrate a practical workflow for the conception, validation, and application of ML-PESs.
• Investigate spectroscopic properties and free energy landscapes for para-chloro-phenol in gas and condensed phases.

Main Methods:

• Development and integration of the MLpot extension using the PhysNet ML model within pyCHARMM.
• Application of the ML-PES to simulate para-chloro-phenol in both gas and aqueous phases.
• Calculation of infrared (IR) spectra and analysis of the free energy for -OH torsion.

Main Results:

• Computed IR spectra for para-chloro-phenol in water show good qualitative agreement with experimental data.
• Relative intensities of spectral features are consistent with experimental findings.
• The rotational barrier of the -OH group increases in water due to hydrogen bonding, from ~3.5 kcal/mol in the gas phase to ~4.1 kcal/mol in solution.

Conclusions:

• The MLpot extension within pyCHARMM provides a robust tool for accurate molecular simulations.
• ML-based PESs effectively capture solvent effects, such as hydrogen bonding, influencing molecular properties.
• This approach facilitates the prediction of spectroscopic observables and reaction dynamics with high fidelity.