Chalmers University of Technology Shenzhen, Guangdong, China (People's Republic)
Enzyme turnover numbers (kcat) are key to understanding cellular metabolism, proteome allocation and physiological diversity, but experimentally measured kcat data are sparse and noisy. Here, we have provided a deep learning approach (DLKcat) to predict kcat values for metabolic enzymes from any organism using only substrate structures and protein sequences. DLKcat can also capture changes in kcat values for mutated enzymes and identify specific amino acid residues that strongly affect kcat values. To facilitate the use of our approach by other researchers, we have compiled a publicly available database called GotEnzymes, which includes predicted turnover numbers and optimal temperature values for all metabolic enzymes in the database across 8099 organisms, provides a valuable resource for researchers in both experimental and computational fields who work with candidate enzymes. Furthermore, we also identified that using the predicted kcat values to parameterize enzyme-constrained genome-scale metabolic models using a Bayesian pipeline can significantly improve the phenotype prediction of these models.