Inference-based assessment of parameter identifiability in nonlinear biological models

Area of Science:

• Systems biology
• Computational biology
• Mathematical modeling

Background:

• Inferring parameter values for biological models from experimental data is crucial for model development.
• Nonlinear models often present challenges in parameter fitting, leading to unidentifiable parameters that cannot be constrained within plausible physiological ranges.

Purpose of the Study:

• To investigate the causes of parameter unidentifiability in biological models.
• To compare different inference-based methods for analyzing parameter uncertainty.
• To provide insights for designing informative experiments.

Main Methods:

• Utilized inference-based methods to analyze parameter unidentifiability.
• Compared a measure-theoretic approach to inverse sensitivity analysis with Markov chain Monte Carlo (MCMC) and approximate Bayesian computation (ABC) for Bayesian inference.
• Mapped output space uncertainty to parameter space probability distributions.

Main Results:

• Identified parameter compensation as a key cause of unidentifiability.
• Demonstrated how the geometry of parameter probability sets reveals sources of unidentifiability.
• Showcased the utility of these methods in guiding experimental design.

Conclusions:

• Parameter compensation significantly contributes to unidentifiability in biological models.
• Inference-based methods, including measure-theoretic approaches, effectively diagnose and visualize parameter unidentifiability.
• The insights gained can inform more efficient and effective experimental design for parameter estimation.