On the identifiability of metabolic network models

Area of Science:

• Systems Biology
• Metabolic Engineering
• Computational Biology

Background:

• Parameter identifiability is crucial for accurate metabolic network modeling.
• Model structure and data limitations often cause identifiability problems.
• Linlog models offer a simplified approach to parameter estimation in metabolism.

Purpose of the Study:

• To develop precise definitions and methods for detecting structural and practical identifiability in linlog metabolic models.
• To adapt identifiability analysis for real-world data scarcity, incompleteness, and noise.
• To apply these methods to assess parameter identifiability in the central carbon metabolism of E. coli.

Main Methods:

• Precise definitions of structural and practical identifiability.
• Singular value decomposition (SVD) for identifiability detection.
• Principal component analysis (PCA) for model reduction to an identifiable approximation.
• Adaptation of criteria for scarce, incomplete, and noisy real data.

Main Results:

• A novel criterion for identifiability analysis adapted for real data was developed.
• Simulated data testing confirmed the criterion's ability to identify principal data components.
• Application to E. coli central carbon metabolism showed only 4/31 reactions and 37/100 parameters were identifiable.
• This highlights the critical need for identifiability analysis and model reduction in large-scale metabolic networks.

Conclusions:

• The developed methods effectively detect and resolve parameter identifiability issues in linlog models.
• Identifiability analysis is essential for reliable metabolic network modeling, especially with limited data.
• The approach is applicable to other pseudo-linear models and offers insights for nonlinear models.