Prediction of Charged Small Molecule Conformations in Solution Using a Balanced ML/MM Potential.

Area of Science:

• Computational chemistry
• Drug design
• Molecular modeling

Background:

• Accurately predicting charged molecule conformations in solution is difficult for computational chemistry.
• Existing methods have accuracy or efficiency limitations.

Purpose of the Study:

• To develop a novel hybrid machine learning/molecular mechanics (ML/MM) potential for accurate prediction of charged small organic molecule conformations in solution.
• To address challenges in drug design related to molecular shape and stability.

Main Methods:

• Utilized metadynamics-enhanced molecular dynamics simulations.
• Developed a hybrid ML/MM potential (PairFEQ-Net) with dynamic, conformationally dependent charges.
• Incorporated a polarizable SWM3 model for water and an empirical charge scaling parameter.

Main Results:

• A basic ML/MM approach showed inferior results compared to fixed-charge models.
• The refined ML/MM potential with dynamic charges achieved mean absolute errors of 0.5 kcal mol⁻¹ in conformational free energy predictions.
• The new potential balances intra- and intermolecular interactions effectively.

Conclusions:

• The developed ML/MM potential provides a chemically accurate method for predicting the solution-phase shapes of charged organic molecules.
• This approach offers a promising route for improving drug design and molecular modeling accuracy.