Synthetic chemists typically run one set of experiments to find conditions that maximise yield, then a separate campaign of time-course measurements to understand why a reaction works. Chemists at the University of Tokyo have collapsed the two into one. Their method, named CYAN (concentration-dependent yield analysis), uses machine learning to fill gaps in routine optimisation data, then applies rate equations from a proposed mechanism to estimate rate constants, so a single dataset serves both discovery and understanding.
As a test case the team studied a nickel-mediated reaction that assembles large ring-shaped carbon molecules. CYAN revealed that nickel accelerates formation of the desired ring while slowing a competing side reaction, favouring the target molecule over a mixture. The researchers describe this as a counterintuitive template effect, since templates are usually assumed only to accelerate chemistry rather than to suppress unwanted pathways.
In a striking demonstration of reach, the group applied CYAN to experiments reported 176 years ago and recovered a rate constant matching the modern value. The work, published in Advanced Science, does not replace precise kinetic experiments, but it offers a practical route to mechanistic insight from existing datasets, including the vast archive of published optimisation tables, and could sharpen how new synthetic methods are designed.