In an article by Foteini Michalopoulou (Imperial College London), the potential of AI-driven surrogate models to enhance real-time control and optimization in downstream bioprocessing is explored—specifically in the context of ion exchange chromatography for monoclonal antibody purification.
Traditional mechanistic models, based on first-principles, are powerful but often too computationally intensive for real-time use. To bridge the gap between accuracy and speed, the study introduces surrogate modeling—using both hybrid models (combining mechanistic insights with data) and fully data-driven models—to accelerate decision-making.
The most important findings from a case study conducted in collaboration with Dr. Maria Papathanasiou:
- Hybrid models outperformed mechanistic models for cycle-to-cycle optimization, improving yield (82% vs. 78%) while maintaining purity standards.
- Data-driven models were best for minute-by-minute updates, offering near-instant predictions and enabling rapid optimization without compromising product quality.
- ·Mechanistic models were too slow for real-time use, especially under tight process windows.
Both surrogate approaches integrate well into digital twins and PAT frameworks, offering virtual sensing capabilities and real-time support when direct measurement is delayed or unavailable. This adaptability makes them valuable tools for Quality by Design (QbD) strategies and agile manufacturing environments.
Ultimately, the article argues for a layered modeling strategy—using mechanistic, hybrid, and data-driven models where each is most effective—to align deep process understanding with the responsiveness demanded by modern biomanufacturing.
Bron: BioprocessOnline