Advances in Machine Learning Models for Predicting Enzyme Kinetic Parameters

Enzyme kinetic parameters, including k_cat, K_m, k_cat/K_m, and K_i, are critical for guiding applications in enzyme engineering, metabolic modeling, and synthetic biology by providing quantitative information on enzyme activity under various conditions. Experimental determination of these parameters is often costly and time-consuming. Moreover, traditional computational methods are not well-suited to estimating these parameters. This motivated the development of machine learning (ML) models for in silico predictions. Here, we review recent advances in ML-based prediction of enzyme kinetic parameters by highlighting global models trained on diverse enzyme classes and local models catered toward specific enzyme families. These models have been applied in myriads of applications including predicting mutation effects, accelerating enzyme mining, and parametrizing genome-scale metabolic models. While data scarcity remains the main limitation for these models, we outline emerging opportunities such as high-throughput data generation and semisupervised learning as means to overcome this issue. In summary, this Review provides a roadmap for leveraging ML to enhance the performance, robustness, and scope of enzyme kinetic parameter prediction, leading to the accurate annotation of protein sequences for target functions.