Identifying and synthesizing molecules that exhibit the desired function is a core and essential component of addressing many contemporary problems in chemistry. For many years, the typical discovery paradigm has been an iterative process of designing candidate molecules, synthesizing those molecules, and then testing their performance. In recent years there has been an increase in the drive to automate the prediction of proposed candidate reaction mechanisms using in silico, data-driven methods. In this study, we present a computational protocol that maximizes the efficiency of the search for elementary reactions by using a stochastic graph-based approach, with limited human intervention, thus reducing the need for large-scale trial and error laboratory synthesis. As an alternative, we present a hybrid approach combining machine learning and electronic structure calculations, supported by a fully developed workflow that relies only on structure information for reactants and products to predict activation barriers of multi-step reaction mechanisms, thereby enabling the identification of the optimal reaction pathway. This work shows the successful application of our approach to the investigation of the formation of benzene in the interstellar medium based on previous literature findings, and find that our decision-making algorithm emerges as the clear winner.

High-throughput screening of mechanistic hypotheses using artificial neural networks and multi-criteria decision making

ACS Spring 2022

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