Chemists at the Massachusetts Institute of Technology have developed a computational model that can quickly ** the structure of the transition state of the reaction (left structure) if given the structure of the reactant (middle) and the product (right). **d**id w. kastner
During a chemical reaction, the molecules gain energy until they reach the so-called transition state – the reaction must start from this point of no return. This state is so ephemeral that it is almost impossible to observe it experimentally.
The structure of these transition states can be calculated using quantum chemistry-based techniques, but this process is very time-consuming. A team of researchers at the Massachusetts Institute of Technology has now developed an alternative method based on machine Xi that can calculate these structures faster in a matter of seconds.
Their new models can be used to help chemists design new reactions and catalysts to produce useful products, such as fuels or drugs, or to mimic naturally occurring chemical reactions, such as those that may help advance the evolution of life on Earth.
Understanding the structure of the transition state is important as a starting point for thinking about designing catalysts or understanding how natural systems implement certain transitions," said Heather Kulik, associate professor of chemistry and chemical engineering at MIT and senior author of the study.
Dr. Chenru Duan is the lead author describing the work, which was published today in Nature Computing Science. Yuanqi Du, a graduate student at Cornell University, and Haojun Jia, a graduate student at the Massachusetts Institute of Technology, are also the authors of the **.
For any given chemical reaction to occur, it must go through a transitional state that occurs when it reaches the energy threshold required for the reaction to proceed. The probability of any chemical reaction occurring depends to some extent on the likelihood of transition state formation.
Transition states help determine the likelihood of a chemical transition occurring. If we have a lot of unwanted things, such as carbon dioxide, and we want to convert it into a useful fuel, such as methanol, then the transition state and how favorable it is determines our likelihood of going from reactant to product," Kulik said.
Chemists can calculate transition states using a quantum chemistry method called density functional theory. However, this approach requires a lot of computing power and can take hours or even days to figure out a transition state.
Recently, some researchers have attempted to discover transitional state structures using machine-based Xi models. However, the models developed so far require treating the two reactants as a single entity, in which the reactants maintain the same orientation to each other. Any other possible orientations would have to be modeled as separate reactions, which would increase the computation time.
If the reactant molecules are spinning, then in principle, they can still undergo the same chemical reaction both before and after this rotation. But in traditional machine Xi methods, the model treats it as two different reactions. This makes machine Xi training more difficult and less accurate," Duan said.
The MIT team has developed a new computational method that allows them to use a model called a diffusion model, which represents two reactants in any arbitrary direction, which can learn Xi which types of processes are most likely to produce specific results. As training data for their model, the researchers used the structures of reactants, products, and transition states computed using quantum computing methods for 9,000 different chemical reactions.
Once the model understands the basic distribution of how these three structures coexist, we can give it new reactants and products, and it will try to produce transitional state structures that are paired with those reactants and products," Duan said.
The researchers tested their model on about 1,000 previously unseen reactions, asking it to generate 40 possible solutions for each transition state. They then use a "confidence model" to determine which states are most likely to occur. Compared to the transitional state structures generated using quantum technology, the accuracy of these solutions is in the 008 angstroms (1/100,000,000,000 centimeters). The entire calculation process for each reaction takes only a few seconds.
As you can imagine, this can really be extended to consider generating thousands of transition states in a time when traditional approaches would normally only need to generate a few transition states," Kulik said.
Although the researchers trained their model primarily on reactions involving compounds with a relatively small number of atoms (up to 23 atoms per system), they found that it could also be accurate for reactions involving larger molecules**.
Even if you look at larger systems or systems catalyzed by enzymes, you can do a good job of covering the different types of ways in which atoms are most likely to rearrange," Kulik said.
The researchers now plan to expand their model to include other components such as catalysts, which could help them investigate the extent to which a particular catalyst accelerates the reaction. This is useful for developing new processes for the generation of drugs, fuels, or other useful compounds, especially when synthesis involves many chemical steps.
Traditionally, all of these calculations have been done with quantum chemistry, and now we are able to replace the quantum chemistry part with this rapidly generative model," Duan said.
Another potential application of such a model, the researchers say, is to explore possible interactions between gases found on other planets, or to mimic simple reactions that may occur during the early evolution of life on Earth.
More information: Generating Accurate Transition States Using Object-Aware Isotropic Fundamental Reaction Diffusion Models, Nature Computational Science (2023). doi： 10.1038/s43588-023-00563-7