During a chemical reaction, molecules move towards a transition state, a high-energy state that dictates how the reaction will proceed. However, this transition state is difficult to predict and observe due to its fleeting nature. Traditionally, scientists use quantum chemistry methods like density functional theory to evaluate these transition states, though these calculations tend to be time-consuming and computationally intensive.
Researchers at the Massachusetts Institute of Technology (MIT) have found a way around these limitations by developing a computational model that employs machine learning. This model estimates the structures of transition states more swiftly and effectively, within a matter of seconds. This capability could assist chemists in designing new reactions and catalysts for the production of crucial materials such as fuels or pharmaceutical drugs or even the modeling of naturally occurring chemical reactions, like those involved in the evolution of life on earth.
“This problem is holding back many important fields such as computational catalyst and reaction discovery, and this is the first paper I have seen that could remove this bottleneck,” says Jan Halborg Jensen, a professor of chemistry at the University of Copenhagen, who was not involved in the research.
The machine learning model employs a technique called diffusion model, which can assess processes are most likely to lead to a specific result. It was trained on approximately 9,000 different chemical reactions, using data derived from quantum computation methods. The model was tested on 1,000 reactions not previously encountered and performed capably, predicting reactions to within 0.08 angstroms of accuracy—the equivalent of one hundred-millionth of a centimeter. All the calculations were produced within seconds.
The approach, developed by MIT graduate student Haojun Jia and Cornell graduate student Yuanqi Du along with their lead author Chenru Duan, also worked well with larger molecules, laying the groundwork for further development in the model to include catalysts. By doing so, the team aims to explore how much a catalyst could hasten a reaction—information which could be invaluable in generating pharmaceuticals, fuels, and other compounds.
The researcher’s groundbreaking work, funded by the U.S. Office of Naval Research and the National Science Foundation, proposes a significant improvement in predicting chemical reactivity, overcoming a considerable obstacle in computational chemistry. This machine-learning model not only accelerates the computations quantum methods take but prevents unnecessary computational work by recognizing that molecules in different orientations can exhibit the same chemical reaction. The model offers a faster, more accurate, and efficient means of studying and predicting complex chemical reactions, with numerous potential applications, from drug creation to the study of extraterrestrial gases.