A team of MIT scientists has developed a machine learning-based model to calculate transition states during chemical reactions, a process which normally requires quantum computing and can take hours or even days to complete. Transition states, which inevitably occur during reactions when molecules reach a particular energy threshold, were previously calculated through quantum chemistry’s density functional theory. However, the researchers’ new model, which constructs a representation of two reactants in an arbitrary orientation relative to each other, is said to be a more efficient and accurate computational approach. Importantly, the new model can generate a transition state structure compatible with those reactants and products in a matter of seconds.
The computational method could be used to understand and design new catalysts, reactions or to model naturally occurring chemical reactions. The researchers used 9,000 different chemical reactions as training data, with structures of reactants, products, and transition states that had been calculated using quantum computation methods. The model was found to be accurate to within 0.08 angstroms, equivalent to one hundred-millionth of a centimeter, when compared to transition state structures generated using quantum techniques.
While the MIT team initially focused on reactions involving relatively small compounds, they discovered that the model can also accurately predict reactions involving larger molecules. Consequently, the researchers aim to enlarge their model’s capabilities to include other components, such as catalysts, allowing them to gauge a particular catalyst’s potential to expedite a reaction. This may prove beneficial to pharmaceutical production or the creation of useful compounds in the future.
Notwithstanding its applications for industry, the model could also be used in the field of astrobiology to understand potential chemical interactions on other planets or even to simulate the early development of life on Earth. The study was applauded by Jan Halborg Jensen, Chemistry professor at the University of Copenhagen, who acknowledged its contribution to predicting chemical reactivity, an area of research previously beset by computational limitations.
