emle-engine: A Flexible Electrostatic Machine Learning Embedding Package for Multiscale Molecular Dynamics...

Area of Science:

• Computational Chemistry
• Molecular Dynamics Simulations
• Machine Learning in Chemistry

Background:

• Hybrid machine learning potential/molecular-mechanics (ML/MM) simulations are crucial for modeling complex chemical systems.
• Accurate representation of electronic charge distribution and induction effects is essential for reliable simulation results.
• Existing methods often struggle with systems exhibiting significant charge variations in the machine learning subsystem or environment.

Purpose of the Study:

• To present the emle-engine package, implementing a novel electrostatic machine learning embedding (EMLE) scheme for ML/MM dynamics.
• To evaluate the performance and stability of the EMLE scheme in enhanced sampling molecular dynamics simulations.
• To demonstrate the superiority of EMLE compared to traditional molecular mechanics (MM) embedding for accurate free energy calculations.

Main Methods:

• Developed the emle-engine package based on a physics-informed model of electronic density and induction.
• The EMLE scheme utilizes tunable parameters derived from in vacuo properties and requires only atomic positions and partial charges.
• Tested EMLE by calculating free energy surfaces of alanine dipeptide in water using various ML potentials and embedding models.

Main Results:

• The EMLE embedding scheme demonstrated stability in enhanced sampling molecular dynamics simulations.
• EMLE significantly outperformed traditional MM embedding with fixed partial charges when compared to DFT/MM reference calculations.
• The inclusion of configurational electronic density dependence and induction energy in EMLE led to a systematic reduction in free energy surface errors.

Conclusions:

• The emle-engine package provides a robust and accurate electrostatic embedding scheme for ML/MM simulations.
• EMLE enables accurate modeling of systems with dynamic charge distributions, advancing the capabilities of computational chemistry.
• This work facilitates the application of advanced ML/MM techniques to complex chemical and biological processes.