Chemists routinely optimize reactions to maximize the yield of their desired products, but understanding why those reactions work can require laborious experiments that track reactions through time. Researchers at the University of Tokyo have developed a method to extract hidden information on reaction speeds from yield data obtained during reaction optimization using machine learning together with rate equations developed by chemists. Concentration-dependent yield analysis (CYAN) uses machine learning to expand the yield data from reaction optimization experiments, then uses this information to calculate how quickly the different steps of a reaction occur. This gives chemists a tool to understand, improve and design complex high-yield chemical reactions.
In order to make a desired molecule, chemists carry out experiments to find the best conditions, changing variables such as concentration, temperature or the times at which ingredients are added. These optimization experiments reveal which conditions achieve the highest yield and can sometimes provide clues about why a reaction works. However, understanding reaction mechanisms in detail has traditionally required separate kinetic experiments that follow how a reaction changes over time. CYAN connects these approaches by extracting kinetic information from the yield data generated during reaction optimization.
“For years, chemists have treated these as two different objectives. We optimize reactions to achieve the highest yield, and if we want to understand the mechanism, we perform another series of experiments that follows the reaction over time. Our goal with CYAN was to bring those two activities together,” said Professor Hiroyuki Isobe from the Department of Chemistry. “With CYAN, machine learning first augments the yield data obtained from our experiments. Chemists then apply rate equations based on their hypotheses about the reaction mechanism to extract rate constants from these data. In this way, a single set of experiments can be used for both reaction optimization and kinetic analysis.”
The researchers developed CYAN by combining machine learning with mathematical descriptions of how they think a reaction works. Machine learning fills in the gaps between the experimental results, creating a more complete picture of how the amount of product changes under different conditions. The researchers then use this information to estimate the speed of different steps in the reaction, without needing separate experiments that measure the reaction repeatedly over time.
“Our interest in this project actually began out of frustration. We had previously developed a machine-learning tool that was very good at finding better reaction conditions, but it behaved like a black box, telling us what worked without telling us why,” said Isobe. “As organic chemists, understanding the chemistry is just as important as improving efficiency. CYAN grew from our attempt to interpret those machine-learning results. We see it as a two-way collaboration between machine learning and organic chemists, in which we contribute the chemical hypotheses and experimental data, while machine learning contributes a wealth of augmented yield data.”
To test the method, the team studied a nickel-mediated reaction used to build large ring-shaped carbon molecules. CYAN successfully extracted kinetic information from the optimization data, uncovering an unexpected feature of the chemical reaction process. Nickel served to accelerate the formation of the desired molecular ring while also slowing down a secondary reaction that would otherwise produce unwanted products. That combination helped steer the reaction toward making a single target molecule rather than a mixture of different ones.
“The nickel result surprised us because it slowed part of the reaction, and that sounds like it should make synthesis worse, not better. A useful way to imagine it is water flowing through a river that splits into two channels. If one channel becomes narrower, more water naturally follows the other route,” said Isobe. “We believe nickel creates a similar effect by restricting one competing pathway, allowing the reaction to concentrate its effort on producing the desired ring-shaped molecule. The “template effect” is normally conceived as an effect that accelerates a reaction, but the effect found in this study was counterintuitive: the template instead retarded part of the reaction. That insight may prove useful for designing many other metal-templated reactions.”
The researchers emphasize that CYAN is intended to complement, not replace, traditional kinetic experiments. When highly precise measurements are required, time-course analysis remains essential. However, CYAN offers a practical way to extract kinetic and mechanistic information from yield data obtained during reaction optimization, potentially helping researchers design new synthetic methods while making better use of both new and previously published experimental datasets. The researchers also found that CYAN could be applied to historical experimental data, suggesting that countless published datasets could potentially be reexamined to reveal underlying chemical kinetics. In one case, the researchers even used data from chemistry experiments reported 176 years ago and obtained a rate constant consistent with a modern value.