Researchers from MIT have used machine learning to expedite the calculation of transient molecular states that occur during chemical reactions. The team’s innovative new model streamlines the process, from a previously time-consuming task, performed using quantum chemistry techniques, to a few seconds. Applied, it could assist chemists to design new reactions and catalysts to create beneficial outputs, such as fuels or drugs, or to understand chemical reactions, such as those that could have influenced the evolution of life. The MIT model can depict two reactants in any orientation to each other. It is based on a diffusion model, enabling it to form an understanding of the processes most likely to result in a specific outcome. Drawing on data from 9,000 distinct chemical reactions, the model was tested on a further 1,000 reactions, and produced accurate results rapidly. Plans exist to refine the model further by incorporating greater variables, such as catalysts. One expert praised the new method as a “significant step forward in predicting chemical reactivity”.
Skip to content
Skip to footer
