Seminars

Home Centers & Programs AI and Natural Sciences Seminars

TITLE: Neural Network-Driven Featurization of Chemical Structures: From Microscopic Interactions to Macroscopic Properties

ABSTRACT: Intermolecular and intramolecular interactions lie at the core of numerous chemical and biological processes. The electrostatic potential between electrons and nuclei governs those molecular interactions, the formation of ordered structures and collective dynamics at the molecular level, and the macroscopic properties of various substances—both at equilibrium and nonequilibrium conditions. Despite the enormous potential of machine learning in computational chemistry, the limited availability of experimental data became a significant challenge, particularly in the face of increasingly complex neural network models. We propose a novel neural network-based approach that efficiently derives high-fidelity molecular features from rich, computer simulation-generated data to extract the molecular interactions and eventually predict various macroscopic properties of the chemical system.