Model parameter estimation and analysis: understanding parametric structure.

Abstract

We developed three algorithms to facilitate an analysis of the parameter combinations (PASS points) that fit experimental data to a desired degree of accuracy. The clustering algorithm separates PASS points into clusters (PASS clusters) as a preliminary step for the following geometrical parametric analyses. The PASS region reconstruction algorithm defines the space of a PASS cluster to allow further parametric structural analysis. The feasible parameter space expansion algorithm produces a complete PASS cluster to be used for model predictions to evaluate the effects of variability and uncertainty. These algorithms are demonstrated using two pharmacokinetic models; a single compartment model for procainamide and a three-compartment physiologically based model for benzene. We found a more thorough representation of the parameter space than previously considered. Thus, we obtained model predictions that describe better the variability in population responses. In addition, we also parametrically identified a subpopulation that may have a higher risk for cancer.