A Machine Learning-Empowered Kinetic Reconstruction of E. coli Metabolism

Constraint-based models of metabolism have been powerful tools for characterizing and predicting organism function with minimal parameter requirements. However, several important applications of metabolic models have emerged that depend upon enzyme kinetic parameters, which are scarce in the literature. In this work, we present a genome-scale kinetic reconstruction of the E. coli metabolic network. We curated the literature and existing databases to extract available kinetic data, including enzyme kinetic parameters, kinetic mechanisms, and protein structural information. To address the data scarcity issue, we fill gaps in measured parameters with statistical models for kcat, Km, and Keq trained on literature data. We reconcile potential inconsistencies in data using a nonlinear regression approach that finds rate constants satisfying kinetic data for each enzyme to the extent possible. The resulting kinetic reconstruction should be a valuable resource to the community and empower parameter-dependent systems biology research.