Computational chemistry has experienced an explosion of usage and development in the past few decades. As computational power continues to grow, so does our set of tools to tackle systems that are becoming progressively more complicated. Typically, one must choose a tool that sufficiently describes a system of interest, while minimizing the computational load. This has led to a split between reactive studies, those that allow the bonding topology to change, and non-reactive studies, which rely on a fixed bond topology. QM/MM methods have been developed to bridge the two simulation types so that the researcher may look at large systems and also capture a description of the reactive event. However, these methods suffer several drawbacks. The QM sphere is the most computationally time-consuming piece of the calculation. Having a QM sphere that is too large slows down the simulation. A QM sphere that is too small risks introducing artifacts to the system when a species treated with QM diffuses to the MM sphere. Further, a reaction is a rare event and as such, the simulation has to run sufficiently long for the reaction to occur. Given that a reaction is a rare event, full QM treatment may not be necessary until right before the reaction is to occur. To this end, we are using machine learning to develop a deep neural network that can differentiate between reactant, transition, and product states. To do this, we are developing a database of organic reaction mechanisms and executing molecular dynamic simulations for the reactant, transition, and product states. The simulations are analyzed to determine how the surrounding environment differs between the three states. This information will be fed into a machine-learning algorithm to develop a neural network that can differentiate between different reactive states.

Identifying reactant, transition, and product states as a function of their environment with machine learning

Where thousands of chemistry professionals meet to share ideas and advance scientific and technical knowledge. Spring 2023 abstracts, posters, and presentations will launch at the end of March.

ACS Spring 2023

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