Robust Chemical Reaction Prediction Enabled by Sim-to-Real Transfer and Self-Supervised Learning

One of the major subjects of chemistry is studying chemical reactions, the transformations of matter, and predicting their outcomes is one of the fundamental goals. Due to the unimaginably large number of potential reactions, it is impossible for human experts to predict them reliably. Artificial intelligence promises to solve this complex problem, but the available data has proven to be insufficient for developing reliable algorithms. In this research, scientists will develop a new language for chemical reactions and leverage advanced molecular simulation techniques to facilitate learning chemical reaction prediction by artificial intelligence algorithms.