A perennial challenge in computational chemistry is the process of finding the minimum energy configuration of weakly interacting molecules, or molecules interacting with a surface. Predicting the minimum energy structure is vital for modeling a number of chemical applications, including catalysis, design of targeted dugs, and self-assembling surface coatings. In these problems, we often deal with numerous choices for selecting the initial molecule configuration, and this is particularly challenging if the molecule is flexible. The traditional method of finding the lowest energy structure is to perform calculations starting from many different initial configurations. To be sure of finding the global minimum, a very large number of calculations can be needed, but many of these converge at the same result, which is a waste of computational effort. Here, we present a novel AI-based approach to accelerate computational studies of molecule-surface interactions. Our method is based on an active-learning framework that interactively couples Gaussian Process (GPs) regression, Bayesian inference, and molecular dynamics simulations to accelerate the search for the minimum energy structure. To demonstrate application of the proposed method and verify its’ accuracy, we used the approach to study water molecules on the zinc oxide semi-polar surface, and vanillin on zinc. However, the presented approach can be used to study molecule surface interactions in other material systems. Therefore, it addresses a long-standing challenge in computational chemistry and any other problem in which we deal with a large dataset and overlap of the targets.

Bayesian inference approach to accelerate studies of surface-molecule interactions

ACS Spring 2022

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.