AI Model Predicts Chemical Reaction’s Turning Point in Under a Second

Chemists designing new chemical reactions often seek a critical piece of information: the reaction’s transition state, the point of no return. Knowing this allows for optimized conditions that ensure the desired reaction occurs. However, traditional methods for predicting this transition state are complex and computationally intensive.

Researchers at MIT have developed a machine-learning model capable of predicting these transition states with high accuracy in under a second. This breakthrough could revolutionize the design of chemical reactions for producing valuable compounds like pharmaceuticals and fuels.

“We’d like 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,” says Heather Kulik, the Lammot du Pont Professor of Chemical Engineering, a professor of chemistry, and the senior author of the new study.

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 published in Nature Machine Intelligence.

Accurate Transition State Prediction

The transition state represents the energy threshold a chemical reaction must overcome to proceed. These states are so brief they are nearly impossible to observe experimentally. Computational techniques based on quantum chemistry can calculate transition state structures, but this requires significant computing power, sometimes taking hours or days for a single calculation.

Kulik explains, “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, Duan, and colleagues introduced a machine-learning strategy to predict reaction transition states. While faster than quantum chemistry techniques, it still required the model to generate roughly 40 structures and run them through a confidence model to determine the most likely states.

The previous model relied on randomly generated guesses for the starting point of the transition state, requiring numerous calculations to reach the final guess. The new model, React-OT, uses a different approach, initiating from an estimate generated by linear interpolation, positioning each atom halfway between its reactant and product positions.

According to Kulik, this linear guess provides a much better starting point than a random guess. The React-OT model requires fewer steps and less time to generate a prediction. The study demonstrated that React-OT could make predictions in approximately 0.4 seconds with only about five steps. These predictions are also about 25 percent more accurate than the previous model.

Duan notes, “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 Applicability

React-OT was trained using a dataset of 9,000 chemical reactions, including reactants, products, and transition states calculated using quantum chemistry methods. The model demonstrated strong performance on reactions held out from the training data. It also performed well on reactions with larger reactants, including those with side chains not directly involved in the reaction.

Kulik emphasizes the importance of this generalization across different system sizes, enabling it to tackle a wide array of chemistry, including polymerization reactions involving large macromolecules.

The researchers are now focused on expanding the model’s capabilities to predict transition states for reactions involving elements such as sulfur, phosphorus, chlorine, silicon, and lithium.

Markus Reiher, a professor of theoretical chemistry at ETH Zurich, who was not involved in the study, commented that quick prediction of 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 created an app to allow other scientists to utilize their approach in designing reactions. With this tool, users can input a reactant and product, and the model will generate the transition state, enabling estimation of the energy barrier and the likelihood of the reaction occurring.