When chemists see how concentrations of substances change, they need to figure out which reactions are behind those changes. It's like trying to reconstruct a recipe from photos of plates at different stages of a meal. Previously, this often involved derivatives — instantaneous rates of concentration change. But derivatives are highly sensitive to noise: a slight jitter in the data turns into false peaks and dips.

The authors propose a different route: instead of derivatives, they use integrals — the accumulated changes in concentration over a time interval. An integral smooths out random fluctuations, much like averaging many photos removes random glares. Based on the integral form of the equations, the algorithm automatically selects possible reaction mechanisms by testing combinations of substances and their reaction orders.

In tests on noisy data, the method reconstructed both reaction rates and the structure of the interaction network more accurately than derivative-based approaches. This means even imperfect laboratory data can lead to the correct reaction scheme. Such an automated tool could be useful in pharmaceuticals, ecology, and anywhere complex chemical processes need to be understood without manually testing hypotheses.