Optimal design for model discrimination using the F-test with non-linear biochemical models. Criteria for choosing the number and spacing of experimental points

Abstract

The problem of choosing the number and spacing of experimental measurements in order to distinguish between several important biochemical models is investigated. Functionals S n (ϑ_), Q (ϑ_) and R ( n ) are introduced as optimal design criteria for model discrimination with multiple ligand binding sites, consecutive compartmental transport and denaturation of independent isoenzymes. Software to calculate these functionals and record their behavior as functions of the true parameter vector ϑ_ and the number of design points n is described. Optimisation over a subset of possible designs suggests that geometrically increasing spacing is to be preferred. This conclusion and conclusions about the number of measurements required are also reinforced by Monte Carlo simulation using the F -test for model discrimination.