Design and Experimental Validation of a Photocatalyst Recommender Based on a Large Language Model

Abstract Utilizing an extensive library of literature on photocatalytic transformations, we disclose the development of a machine learning (ML) model for the recommendation of photocatalysts most suitable for reactions of interest. The model is trained on > 36 000 such literature examples and uses an architecture inspired by the Bidirectional Encoder Representations from Transformer (BERT) large language model. Under cross‐validation, it can suggest the “correct” photocatalysts with ∼90% accuracy. When experimentally tested on five out‐of‐box reactions, this algorithm consistently suggested photocatalysts that gave yields competitive to those chosen by human researchers and frequently suggested alternative photocatalysts that are potentially more appealing than the originally selected photocatalyst. Altogether, this platform serves as a valuable tool for researchers undertaking reaction optimization programs. The model is free to use at https://photocatals.grzybowskigroup.pl/predict/ .