Automated Data-Efficient Symbolic Regression for Interpretable Bioprocess Model Development

Area of Science:

• Biotechnology
• Chemical Engineering
• Computational Biology

Background:

• Bioprocessing is vital for sustainable production of pharmaceuticals, food, and chemicals.
• Accurate kinetic models are essential for process prediction, optimization, and scale-up.
• Model development is challenged by incomplete understanding and limited data.

Purpose of the Study:

• To present a data-efficient symbolic regression (SR) framework for automated bioprocess model construction.
• To simultaneously identify interpretable model structures and facilitate knowledge discovery.
• To accelerate the development of high-fidelity kinetic models for bioprocesses.

Main Methods:

• Employed a generic macroscopic kinetic model backbone.
• Applied symbolic regression (SR) to uncover structures of critical kinetic terms.
• Evaluated two SR implementation strategies: embedded SR and pre-identification of time-varying parameters.
• Utilized a novel local iterative structural correction strategy for refining SR candidates.

Main Results:

• Independently identifying kinetic terms was key to recovering the ground-truth model.
• The local iterative structural correction strategy significantly improved convergence to true kinetic expressions.
• The SR-based framework demonstrated superior data efficiency compared to model-based design of experiments.
• Successfully applied to an in-silico yeast fermentation case study.

Conclusions:

• The developed framework enables automated and interpretable model construction for bioprocesses with limited data.
• This approach facilitates augmented intelligence-driven bioprocess modeling.
• It accelerates the development of digital twins for enhanced process optimization and control.