The transition of the chemical industry towards renewable biocatalysis requires robust kinetic models in order to scale up and to optimise enzymatic processes. Deriving such models usually relies on the manual fitting of predefined equations by trial and error, while approaches based on machine learning often produce accurate but physically meaningless black box models.
In this work, the authors present an adaptive framework for the evaluation of kinetic models that automates the discovery of enzymatic rate laws directly from batch progress curves. By employing Grammar-Guided Symbolic Regression (GGSR), the algorithm constructs rate equations autonomously, using only kinetic building blocks that are physically permissible.
The reliability of the framework was first validated against 13 synthetic kinetic mechanisms, which confirmed structural and parametric identifiability. The pipeline was then applied to an experimental case study, the isomerisation of glucose to fructose with an isomerase from Streptomyces murinus.
The algorithm successfully recovered a reversible kinetic equation that features an apparent interaction term of the uncompetitive type, and it provided specific activities and Michaelis constants that agree well with the literature. The work demonstrates that symbolic regression of this kind can bridge the gap between automation based on data and biochemical engineering models that remain mechanistically interpretable.
The study is a collaboration between the Institute of Process Systems Engineering and the Institute of Technical Microbiology at Hamburg University of Technology.
Thomas Waluga, Jorge Adrian Peña Ayala, Miriam Edel-Teichmann, Mirko Skiborowski (2026). Grammar-Guided Symbolic Regression as Tool for Model Development of Enzymatic Reactions. Digital Chemical Engineering, 100346.
