Integrated hybrid modelling of lignin bioconversion

Area of Science:

• Biotechnology
• Chemical Engineering
• Computational Biology

Background:

• Biomanufacturing expansion faces challenges due to cellular complexity, limiting traditional modeling approaches.
• Phenomenological models lack robustness, while machine learning (ML) models struggle with extrapolation.
• Hybrid models offer a solution by integrating physical constraints with ML flexibility.

Purpose of the Study:

• To develop a novel, integrated hybrid modeling approach for biomanufacturing processes.
• To enhance the robustness and extrapolative capabilities of biomanufacturing models.
• To demonstrate the application and benefits of the integrated approach in a specific bioplastic precursor production scheme.

Main Methods:

• Developed a singular-step method integrating time-variant parameter estimation and ML model training.
• Implemented the integrated hybrid model for cell-mediated conversion of a lignin derivative to a bioplastic precursor.
• Performed interpretability analysis on the ML component to derive physical insights.

Main Results:

• The integrated hybrid model significantly outperformed traditional two-step hybrid, phenomenological, and standalone ML models.
• The interpretability analysis successfully revealed new physical insights into the biomanufacturing process.
• These insights were utilized to further enhance the integrated hybrid model's performance.

Conclusions:

• The integrated hybrid modeling approach offers a superior strategy for optimizing complex biomanufacturing processes.
• This method enhances model performance and provides valuable physical insights for process improvement.
• The approach holds significant potential for advancing the field of industrial biotechnology.