MIT Model Predicts Chemical Reaction’s Turning Point with AI

Chemists designing new reactions rely on understanding the transition state—the critical juncture from which a reaction inevitably proceeds. Knowing this allows for fine-tuning conditions to favor desired outcomes. Traditional methods for predicting transition states are complex and computationally intensive.

Researchers at MIT have introduced a machine-learning model capable of predicting these transition states rapidly and accurately. This innovation promises to streamline the design of chemical reactions for producing pharmaceuticals, fuels, and more.

Heather Kulik, the Lammot du Pont Professor of Chemical Engineering and a professor of chemistry at MIT, emphasizes the importance of this advancement: “We’d like to be able to ultimately design processes to take abundant natural resources and turn them into molecules that we need, such as materials and therapeutic drugs. Computational chemistry is really important for figuring out how to design more sustainable processes to get us from reactants to products.”

The research, led by Chenru Duan PhD ’22 (now at Deep Principle), Guan-Horng Liu (formerly of Georgia Tech, now at Meta), and Yuanqi Du (Cornell University), is detailed in a paper published in Nature Machine Intelligence.

Estimating Transition States

Chemical reactions must surpass a transition state, requiring a specific energy threshold. These states are so brief that direct observation is nearly impossible. Quantum chemistry calculations offer an alternative but demand significant computing resources, potentially taking hours or days for a single transition state.

Kulik notes the challenge: “Ideally, we’d like to be able to use computational chemistry to design more sustainable processes, but this computation in itself is a huge use of energy and resources in finding these transition states.”

In 2023, Kulik and Duan introduced a machine-learning approach to predict transition states faster than quantum chemistry. However, it still required generating approximately 40 structures and using a “confidence model.”

The new model, React-OT, improves upon this by starting with a linear interpolation estimate of the transition state, positioning each atom halfway between its reactant and product locations.

“A linear guess is a good starting point for approximating where that transition state will end up,” Kulik explains. “What the model’s doing is starting from a much better initial guess than just a completely random guess, as in the prior work.”

This approach reduces the steps and time needed for prediction. React-OT achieves predictions in about 0.4 seconds with only five steps, surpassing the accuracy of its predecessor by approximately 25 percent, without needing a confidence model.

Duan highlights the practical implications: “That really makes React-OT a practical model that we can directly integrate to the existing computational workflow in high-throughput screening to generate optimal transition state structures.”

Broad Chemical Application

React-OT was trained on a dataset of 9,000 chemical reactions, encompassing reactants, products, and transition states calculated via quantum chemistry. It demonstrated strong performance on held-out reactions and generalized effectively to reactions with larger reactants.

Kulik emphasizes the model’s versatility: “This is important because there are a lot of polymerization reactions where you have a big macromolecule, but the reaction is occurring in just one part. Having a model that generalizes across different system sizes means that it can tackle a wide array of chemistry.”

Future work aims to extend the model’s capabilities to reactions involving elements like sulfur, phosphorus, chlorine, silicon, and lithium.

Markus Reiher, a professor of theoretical chemistry at ETH Zurich, who was not involved in the study, commented: “To quickly predict transition state structures is key to all chemical understanding. The new approach presented in the paper could very much accelerate our search and optimization processes, bringing us faster to our final result. As a consequence, also less energy will be consumed in these high-performance computing campaigns. Any progress that accelerates this optimization benefits all sorts of computational chemical research.”

The MIT team has made their approach accessible via an app, encouraging other scientists to leverage it in their research.

“Whenever you have a reactant and product, you can put them into the model and it will generate the transition state, from which you can estimate the energy barrier of your intended reaction, and see how likely it is to occur,” Duan says.

Funding for the research was provided by the U.S. Army Research Office, the U.S. Department of Defense Basic Research Office, the U.S. Air Force Office of Scientific Research, the National Science Foundation, and the U.S. Office of Naval Research.