Comparison of physics-based and data-driven modelling techniques for dynamic optimisation of fed-batch bioprocesses

Area of Science:

• Biotechnology
• Process Engineering
• Computational Biology

Background:

• Digital bioprocessing is essential for modern industrial bioprocesses.
• Optimizing long-term bioprocesses requires advanced modeling techniques.
• Microalgal lutein production is a key area in industrial biotechnology.

Purpose of the Study:

• To investigate the efficiency of physics-based versus data-driven models for dynamic bioprocess optimization.
• To compare the predictive accuracy and optimization performance of these models for microalgal lutein production.
• To evaluate the potential of data-driven models in industrial bio-manufacturing.

Main Methods:

• Developed and employed a predictive kinetic (physics-based) model and a data-driven model.
• Utilized fed-batch operation for microalgal lutein production.
• Applied open-loop optimization strategies using light intensity and nitrate inflow rate as control variables.
• Employed various optimization algorithms to compute optimal control sequences.
• Conducted experimental verification to compare model predictions with actual results.

Main Results:

• Physics-based and data-driven models yielded contradictory optimization strategies.
• The data-driven model demonstrated higher predictive accuracy compared to the physics-based model.
• Both models increased intracellular lutein content by over 40%.
• The data-driven model achieved a 40-50% increase in total lutein production, outperforming the kinetic model.

Conclusions:

• Data-driven modeling offers advantages for optimizing and predicting complex dynamic bioprocesses.
• The data-driven approach shows significant potential for industrial bio-manufacturing systems.
• Experimental validation confirmed the superiority of the data-driven model in this specific application.