Abstract
The recently published DLKcat model, a deep learning approach for predicting enzyme turnover numbers (k (cat)), claims to enable high-throughput k (cat) predictions for metabolic enzymes from any organism and to capture k (cat) changes for mutated enzymes. Here, we critically evaluate these claims. We show that for enzymes with <60% sequence identity to the training data DLKcat predictions become worse than simply assuming a constant average k (cat) value for all reactions. Furthermore, DLKcat's ability to predict mutation effects is much weaker than implied, capturing none of the experimentally observed variation across mutants not included in the training data. These findings highlight significant limitations in DLKcat's generalizability and its practical utility for predicting k (cat) values for novel enzyme families or mutants, which are crucial applications in fields such as metabolic modeling.