Statistics Seminar - Professor Dave Woods (University of Southampton)
| Dates: | 4 November 2026 |
|---|
| Times: | 14:00 - 15:00 |
|---|
| What is it: | Seminar |
|---|
| Organiser: | Department of Mathematics |
|---|
| Who is it for: | University staff, External researchers, Current University students |
|---|
|
Title: Optimal design of dynamic experiments with profile variables
Abstract: In many experiments, particularly in biological and pharmaceutical process development, some controllable variables need not be held fixed within a run. The temperature or feeding regime of a bioreactor, for example, can be varied over the course of a reaction. Designing such a dynamic experiment means choosing, for each run, a function of time for each of these profile variables rather than a single setting.
I will describe a Bayesian decision-theoretic framework for this problem under a scalar-on-function linear model, in which an end-of-run response depends on the profile variables through unknown functional parameters. Representing both the profile variables and the functional parameters via basis expansions admits the use of optimal design methods for a standard linear model, with a model matrix available in closed form for B-spline and monomial bases. Physical constraints on the profiles, such as ranges or monotonicity, can be imposed through linear constraints on the basis coefficients. The framework makes explicit the trade-off between the complexity of profiles that can be implemented in the laboratory and the information gained about the functional parameters. Minimising expected squared error losses on either the coefficients or the implied parameter functions gives Bayesian A- and L-optimal designs.
The methods are illustrated through examples exploring how design performance depends on model and profile complexity, and through experiments performed on an Ambr250 bioreactor system at GlaxoSmithKline.
This is joint work with Damianos Michaelides (Cyprus Institute of Neurology and Genetics), Maria Adamou (Circana) and Antony Overstall (University of Southampton).
Travel and Contact Information
Find event
Frank Adams 2
Alan Turing Building
Manchester
