Parameter identifiability analysis and visualization in large-scale kinetic models of biosystems.

Area of Science:

• Systems Biology
• Computational Biology
• Mathematical Modeling

Background:

• Kinetic models in biochemistry often contain numerous parameters, many of which are practically unidentifiable due to insufficient data or parameter interdependence.
• Identifying parameter identifiability is crucial before model calibration to understand model behavior and limitations.

Purpose of the Study:

• To develop a methodology for detecting high-order parameter relationships and visualizing identifiability in biochemical models.
• To facilitate efficient parameter estimation and improve the design of experiments for model refinement.

Main Methods:

• Utilized a collinearity index for efficient quantification of parameter correlations within groups.
• Applied integer optimization to identify large sets of uncorrelated parameters and small sets of highly correlated parameters.
• Developed a visualization tool (Cytoscape compatible) to display parameter identifiability alongside model structure.

Main Results:

• Introduced a MATLAB toolbox, VisId, implementing the proposed identifiability analysis techniques.
• Demonstrated the combination of global optimization and regularization for calibrating large-scale biological models efficiently.
• Evaluated the practical identifiability of estimated parameters using the developed methodology.

Conclusions:

• The developed approach enhances scalability for practical identifiability analysis of large dynamic models.
• Accelerated model calibration and provided visualization tools for identifying problematic model components.
• Aids modelers in refining models and experimentalists in designing informative new experiments.