Using AI to predict enzyme efficiency at genome scale
Researchers have developed a new artificial intelligence approach that can predict how efficiently enzymes work, potentially helping accelerate the discovery and engineering of improved enzymes.
Published in Genome Biology, the research introduces KcatNet, a geometric deep learning model that predicts enzyme turnover, a measure of how quickly an enzyme can carry out a chemical reaction, using information about both the enzyme and the molecule it acts on.
KcatNet consistently outperformed existing prediction methods, particularly for highly efficient enzymes, and was also able to make accurate predictions for enzymes that differed substantially from those used to train the model.
The researchers applied KcatNet across the genomes of diverse yeast species, using its predictions to improve models of how cells allocate their proteins during metabolism. Experimental testing also confirmed that the model could help identify enzyme mutations with increased activity.
By connecting information about protein sequences, structures and function, the research demonstrates how AI can help predict enzyme behaviour at a much larger scale than is practical through experimental testing alone. This could support future work in enzyme engineering, biotechnology and metabolic modelling.
Read the paper: https://link.springer.com/article/10.1186/s13059-026-03986-3