A Hybrid Mechanistic Model Capturing Lactate Run-Away and Back-Metabolism in Mammalian Cell Culture.

Mechanistic modeling is important in biopharmaceutical manufacturing by enabling in-silico experimentation. However, developing a robust mechanistic model for mammalian cell culture remains challenging due to the complexity of the underlying bioprocesses. Existing data-driven models often suffer from limited extrapolation capability and require extensive datasets. Not all modeling approaches can accurately represent differing metabolic scenarios such as lactate consumption and production. To address these limitations, this study proposes a hybrid mechanistic model for mammalian cell culture to improve predictive accuracy and process understanding. Our model integrates high-frequency pH data with low-frequency measurements of viable cell density (VCD), glucose, lactate, monoclonal antibody (mAb) concentrations from 10 runs (7 L). Novel kinetic equations for cell growth and death, and lactate metabolism considering both scenarios were developed. Additionally, our model enables automatic lactate switch detection based on time-series pH and feeding data. It demonstrated high predictive capability with a mean R2 of 0.88 for VCD, mAb, glucose, and lactate concentrations across training runs. Real-time fine-tuning further improved accuracy, reducing the root mean squared error (RMSE) for glucose by ~50% and for lactate by ~30%. The results indicate that our model provides accurate real-time predictions of VCD, mAb, glucose, and lactate concentrations under variable bioprocessing conditions.