Metabolism-Informed Bioreactor Model for Transferring Knowledge from Fed-batch to Continuous Biomanufacturing

A metabolism-informed bioreactor model for CHO-cell mAb production transfers from fed-batch to continuous operation, where fine-tuning outperforms other transfer learning variants.

• Why it matters: Continuous biomanufacturing needs reliable real-time monitoring and prediction, but limited intracellular observability and sparse extracellular data leave a gap between mechanistic understanding and deployable control.
• What they did: The model embeds reactor- and intracellular mass balances into a learning architecture without predefined kinetic rate laws or explicit steady-state assumptions, acting as a soft sensor for intracellular flux inference and one-step-ahead extracellular forecasting.
• The result: Concentration predictions are indistinguishable across two metabolic network sizes, exchange flux fidelity is solver-agnostic, and transferred fine-tuning improves forecasting while predicted fluxes suggest growth rate—not operating mode—drives metabolic variation.

AI summary of the abstract · may contain errors · not medical advice · read the paper.