Parameter subset selection techniques for problems in mathematical biology.

Area of Science:

• Computational Biology
• Mathematical Modeling
• Physiological Systems Analysis

Background:

• Patient-specific models are crucial for diagnostics and treatment planning, but rely on accurate parameter estimation.
• Physiological models, often nonlinear ordinary differential equations, present challenges due to numerous parameters and limited state variable measurements.
• Determining parameter identifiability from available data is a significant hurdle in model development.

Purpose of the Study:

• To investigate practical parameter identifiability challenges in nonlinear ordinary differential equation models.
• To evaluate methods for selecting identifiable parameter subsets for robust model predictions.
• To demonstrate these methods using diverse examples, including a patient-specific arterial blood pressure model.

Main Methods:

• Exploration of parameter identifiability using local sensitivity analysis.
• Application of global sensitivity analysis techniques, including Morris screening.
• Validation through five examples of increasing complexity and a patient-specific physiological model.

Main Results:

• Local sensitivity methods offer computational efficiency and good model fit with known initial parameter values.
• Global methods are recommended when initial parameter values are unknown or uncertain.
• Morris screening provides cost-effective parameter sensitivity ranking for global sensitivity analysis.

Conclusions:

• Careful selection of identifiable parameters is essential for reliable patient-specific model predictions.
• The choice between local and global sensitivity methods depends on the availability and quality of initial parameter estimates.
• Efficient global sensitivity analysis using Morris screening aids in understanding parameter importance.