Multipolar Electrostatic Energy Prediction for all 20 Natural Amino Acids Using Kriging Machine Learning

Area of Science:

• Computational Chemistry
• Machine Learning
• Biophysics

Background:

• Accurate prediction of electrostatic interactions is crucial for understanding molecular behavior.
• Existing methods may struggle with the complexity of amino acid electrostatics.

Purpose of the Study:

• To develop a generic machine learning method for predicting amino acid electrostatic properties.
• To accurately model molecular electrostatic interaction energies based on geometry.

Main Methods:

• Application of kriging, a machine learning technique, to all 20 naturally occurring amino acids.
• Development of kriging models to predict electrostatic multipole moments from molecular geometry.
• Validation using 200 unseen test geometries per amino acid.

Main Results:

• Mean prediction errors below 5.3 kJ/mol for all amino acids, with a global mean error of 4.2 kJ/mol.
• Consistent accuracy for neutral, charged (protonated/deprotonated), and aromatic amino acid systems.
• Demonstrated ability to capture multipolar polarizable electrostatics generically.

Conclusions:

• The proposed kriging method provides a robust and accurate approach for predicting amino acid electrostatics.
• This generic methodology simplifies electrostatic energy calculations in computational chemistry.
• The findings have implications for molecular modeling and drug design.