Determining Partial Charge Distribution of a Molecule through Machine Learning

Abstract

By Amelia Solomon, Chemistry

Advisor: Thomas Beck

Presentation ID: AM_D37

Abstract: A partial charge of a molecule is used to determine the electrostatic potential of the force field surrounding a molecule. Partial charges also help determine the reactivity of a molecule and can provide a general understanding of the structure of the molecule. There are several properties that help determine partial charges: Electron density, dipole moments, or relationships between the physical properties. Molecular dynamic simulations are a convenient method for viewing the physical movements of these properties. Even with these methods, there is always a high level of uncertainty. And the same method will provide varying levels of accuracy for different molecules. Machine learning is a unique way to determine partial charges by letting the machine determine the best partial charge distribution for a molecule. The machine itself will provide a less biased method of charge development by deriving distribution directly from trends in quantum calculations. The goal is to use unsupervised machine learning to determine a learning set, a set of parameters, for the supervised machine learning program. Ideally, once this program is written, it will be able to analyze any size of molecule and determine a highly accurate partial charge distribution.