Analytical Model of Electron Density and Its Machine Learning Inference.

Area of Science:

• Computational Chemistry
• Quantum Chemistry
• Biomolecular Modeling

Background:

• Accurate electron density (ρ(r)) representation is crucial for understanding molecular properties.
• Existing methods for calculating electron density can be computationally expensive, especially for large biomolecules.
• Analytical models offer a computationally efficient alternative but often struggle to capture molecular anisotropy.

Purpose of the Study:

• To develop an analytical model for molecular electron density that accounts for atomic anisotropy.
• To create a computationally efficient method for determining the model parameters, suitable for large biomolecules.
• To enable the prediction of molecular properties directly from the electron density representation.

Main Methods:

• Developed the Anisotropic Analytical Model of Density (A2MD) based on atomic expansions of exponential and Gaussian functions.
• Introduced anisotropic functions to represent the electron distribution of atoms within molecules.
• Implemented a machine learning (ML) approach (A2MDnet) using neural networks to predict A2MD parameters, bypassing costly ab initio calculations.

Main Results:

• The A2MD model provides an analytical representation of electron density with linear computational scaling concerning the number of atoms.
• A2MDnet successfully predicts A2MD model parameters, yielding reliable electron densities comparable to ab initio methods.
• The ML-based approach significantly reduces the computational cost associated with obtaining electron density parameters.

Conclusions:

• The A2MD model offers an efficient and accurate analytical representation of electron density for large biomolecules.
• A2MDnet presents a powerful, computationally inexpensive alternative for parameterizing the A2MD model.
• These advancements facilitate the study of electron density and related molecular properties in complex biological systems.