Systematic data-driven genome-scale metabolic model reduction for dynamic bioprocess modeling: CHO cell culture case study

Genome-scale metabolic models (GEMs) enable mechanistic insight into cellular metabolism, but their size and underdetermination hinder use in dynamic bioprocess simulation and real-time digital twins. Compact models are essential, yet existing reduction strategies either neglect experimental uncertainty, rely on simplistic rate estimates, or depend on manual assumptions, limiting robustness and scalability. Here, we present a metabolomics-driven reduction pipeline that integrates Bayesian flux estimation to propagate uncertainty from noisy and sparse exo-metabolomics data directly into the reduction process. Applied to time-course data from 12 fed-batch CHO cultures, the method produced a single reduced model that remained feasible across all conditions, avoided over- and under-pruning, and accurately reproduced observed extracellular fluxes. Despite relying solely on exo-metabolomics, the reduced model preserved broad metabolic functionality, highlighting the strong predictive power of extracellular data. This establishes a systematic, uncertainty-aware framework for generating compact GEMs suited for dynamic bioprocess simulation and digital twin integration, demonstrated here in a CHO case study but generalizable across cell systems.