A method for robust model order reduction in pharmacokinetics.

Area of Science:

• Computational Biology
• Pharmacokinetics and Pharmacodynamics
• Mathematical Modeling

Background:

• Model reduction simplifies complex systems but often lacks robustness to parameter variations.
• Existing methods for reducing large systems of differential equations may fail when parameter values vary.

Purpose of the Study:

• To develop a Bayesian automated method for model reduction by lumping that robustly handles parameter variability.
• To improve the reliability of model reduction techniques in applications like Physiologically Based Pharmacokinetic (PBPK) and Systems Biology models.

Main Methods:

• A stepwise Bayesian algorithm was developed to determine optimal lumping schemes, incorporating prior parameter distributions.
• The method reduces system dimensionality one step at a time, with each step conditional on the previous ones.
• Applied to a literature-based PBPK model for barbiturates with introduced parameter variability (20% CV).

Main Results:

• The Bayesian method produced a lumping scheme that was optimal on average across the parameter distribution.
• This outperformed a non-Bayesian method, which yielded a scheme optimal only for nominal values but poor otherwise.
• The Bayesian approach demonstrated superior robustness and performance in the presence of parameter variability.

Conclusions:

• The developed Bayesian automated lumping method effectively addresses the lack of robustness in model reduction due to parameter variability.
• Lumping strategies offer a more powerful approach to model reduction compared to state elimination, which can be seen as a special case.
• This method provides a robust tool for simplifying complex biological and pharmacokinetic models, enhancing their practical utility.