Redox potential plays a crucial role in many applications, and accurately estimating it can be time-consuming and resource-intensive. In this study, we present a novel method for fast estimation of redox potential using message passing neural networks (MPNN). By training on an extensive OMEAD dataset, we achieved the lowest mean absolute error (MAE) among existing approaches reported in the literature, establishing our method as state-of-the-art. Additionally, we introduce ReSolved — our own Dataset of reduction potentials in different solvents and extend MPNN’s capability to generalize across different solvents, broadening its applicability in various chemical environments. Furthermore, we combined our MPNN approach with an evolutionary algorithm to explore the vast chemical space for potential good candidates. This improvement marks a crucial step towards accurately predicting redox potentials in diverse conditions, thereby greatly accelerating the discovery of new catalysts and materials for redox reactions. Ultimately, our method contributes to the development of more efficient and sustainable chemical processes by enabling rapid and versatile evaluations of redox potentials.
Rostislav Fedorov