Predictive modelling of CHO cell culture using enzyme-constrained genome-scale metabolic models coupled with empirical regression
Abstract
Mammalian genome-scale metabolic models (GEMs) are underdetermined due to thousands of reactions but limited constraints, often limiting their predictive accuracy, especially when data are scarce. In this study, we developed a curated pipeline that integrates enzyme-constrained GEMs (ecGEMs) with flux balance analysis (FBA) and regression-based, biologically motivated constraints. Using batch datasets with 21 samples, non-growth associated maintenance ( NGAM ) and the upper bound of glutamate dehydrogenase flux ( G L U D UB ) were parameterised as Hill functions of lactate and ammonia concentrations and imposed on the ecGEM to refine phenotype predictions. Transitioning from the base GEM to the ecGEM substantially improved accuracy by capturing proteome allocation limits, while incorporating NGAM further enhanced growth predictions and adding G L U D UB corrected for overestimated ammonia flux through feedback inhibition. However, these improvements came at a cost of lactate predictions, as restricting GLUD reduced NADH generation and diverted redox balance toward oxidative phosphorylation rather than lactate overflow. Together, the results highlight the trade-offs inherent in constraint-based modelling: each new constraint improves fidelity in one dimension but may impair another. This work demonstrates the value of combining enzyme and stress-dependent constraints to recapitulate overflow metabolism and improve predictive accuracy, while motivating future refinements with omics data and cell-line-specific biomass composition for model-guided rational culture optimisation.
Authors: Hazel Wee, Sizhe Qiu, Claudio Angione, Hua Ye, Akinlolu Oyekunle Oluseun Odeleye, Aidong Yang
Published in: Biochemical Engineering Journal (2026)