Fast and accurate generation of ab initio quality atomic charges using nonparametric statistical regression

Area of Science:

• Computational Chemistry
• Molecular Modeling
• Quantum Chemistry

Background:

• Accurate partial atomic charges are crucial for describing molecular electrostatics in bioorganic systems.
• Existing methods often lack the flexibility to capture complex charge distributions.
• Need for transferable and consistent charge models for various chemical environments.

Purpose of the Study:

• To develop a novel partial atomic charge assignment method with ab initio quality.
• To create flexible models capable of detailed charge distribution analysis.
• To improve the accuracy and transferability of partial atomic charges for bioorganic molecules.

Main Methods:

• Utilized Random Forest regression to build element-specific charge models (H, C, N, O, F, S, Cl).
• Trained models on partial charges derived from b3lyp/6-31G* electrostatic potential (ESP) fitting.
• Refined charges for enhanced consistency and transferability.

Main Results:

• Achieved low mean unsigned error (approx. 1 kcal/mol) for hydration free energy calculations.
• Accurately reproduced electrostatic potential (ESP) and dipole moments from ab initio calculations on diverse organic molecules.
• Developed a method that automatically estimates potential errors in charge assignments.

Conclusions:

• The new method provides a robust and accurate description of electrostatics in bioorganic molecules.
• The approach offers flexibility and detail in capturing charge distribution complexities.
• This work paves the way for improved charge models and methods for predicting chemical properties.