Enhancing Electrostatic Embedding for ML/MM Free Energy Calculations.

Area of Science:

• Computational chemistry
• Molecular modeling
• Machine learning applications

Background:

• Hybrid ML/MM methods balance computational cost and accuracy in simulations.
• Current ML/MM simulations often use mechanical embedding and simplified intermolecular potentials.
• Electrostatic embedding offers improved accuracy by including polarization effects.

Purpose of the Study:

• To develop and validate the Electrostatic Machine Learning Embedding (EMLE) method for hybrid ML/MM simulations.
• To establish robust training methodologies for EMLE models using quantum mechanical data.
• To assess the accuracy of EMLE in modeling electrostatic interactions for organic molecules.

Main Methods:

• Computation of absolute hydration free energies for small organic molecules.
• Development of protocols for fine-tuning static and induced electrostatic components.
• Evaluation of fitting accuracy to first-principles calculations.
• Introduction of an empirical adjustment for experimental agreement.

Main Results:

• Robust methodologies for training EMLE models were derived.
• Accuracy limits of fitting electrostatic components were evaluated.
• An empirical adjustment improved agreement with experimental hydration free energies.
• EMLE strengthens the competitiveness of ML/MM simulations.

Conclusions:

• Electrostatic embedding, via EMLE, enhances ML/MM simulations.
• EMLE provides strategies for accurate modeling of drug-like molecules.
• This approach addresses limitations of conventional MM force fields in specific applications.