Predictive modelling of CHO cell culture using enzyme-constrained genome-scale metabolic models coupled with empirical regression

Accurate prediction of mammalian cell metabolism is complicated by inhibitory by-products such as lactate and ammonia, which are not captured by standard genome-scale metabolic models (GEMs). We present a process-aware modelling framework that extends enzyme-constrained GEMs (ecGEMs) with empirically derived, stress-dependent constraints to improve phenotype prediction under dynamic culture conditions. A reduced ecGEM was constructed from a published Chinese Hamster Ovary model by retaining central carbon and nitrogen metabolism while removing peripheral reactions, improving tractability without loss of key phenotypes. Using an existing cell culture dataset, non-growth-associated maintenance and glutamate dehydrogenase capacity were parameterised as Hill-type functions of extracellular lactate and ammonia to capture metabolic stress responses. The resulting model substantially improved growth rate and ammonia flux predictions under inhibitory conditions, while reproducing trends in lactate overflow. Integration with dynamic flux balance analysis enabled simulation of batch and fed-batch cultures, capturing industrially relevant behaviours including reduced ammonia accumulation and extended culture longevity under low nutrient setpoint control. Together, this framework demonstrates how combining enzyme constraints, stress-responsive regulation and dynamic simulation can enhance the predictive power of metabolic models for mammalian bioprocess optimisation.