Optimal Design of Challenge-Response Experiments for Mechanistic Model Discrimination

• Activity Number: 567
• Type: Contributed
• Date/Time: Wednesday, August 1, 2012 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistical Computing
• Author(s): Matthew Shotwell*+
• Companies: Vanderbilt University
• Keywords: optimal experimental design ; differential equation ; bootstrap ; model discrimination ; preliminary data ; mechanistic model

Abstract

We consider an experimental approach where a system under study is subjected to controlled challenges, with the expectation that responses to these challenges will be informative about the underlying mechanism. For example, we have examined the association of cardiac electrophysiology and metabolism by observing the time course of conduction velocity under intermittent anoxia.

Given two competing mechanistic models and preliminary data, we consider an optimal design strategy to discriminate between mechanisms. Design of challenge-response experiments is unique because the types and durations of challenges modify the responses predicted by competing models. These 'active' design aspects may then be selected such that predicted responses are most divergent among the competing mechanisms. Our approach draws on existing bootstrap methods, and builds on the so-called T-optimality criterion for design of model discrimination experiments.

In addition, we address the notion that although a proposed experiment is optimal, it may be unlikely to yield a result that is discriminative between competing mechanisms. This problem is mostly neglected in the Design of Experiments literature.