Interests in Mathematical/Statistical Biology

My PhD relates to the use of flow-cell optical biosensors to indirectly study the dynamics of biomolecular interactions. This led to interest in variety of problems.

My work feels like this sometimes...

Figure 1: My work feels like this sometimes…

Structural global identifiability (SGI)?

This concept lurks wherever we wish to “calibrate” a model structure (estimate parameters from data) so that we can proceed to make predictions.

Often we can represent a physical system by a structure of “state-space” systems. These relate state variables (x), observables (y) and possibly inputs (u) through parametric expressions in the parameters \(\boldsymbol{\theta}\): \[\begin{gather} \dot{\bf x}({\bf x}, {\bf u}, t; {\boldsymbol{\theta}}) = {\bf f}({\bf x}, {\bf u}, t; {\boldsymbol{\theta}}) \, , \quad {\bf x}(0) = {\bf x_{0}}(\boldsymbol{\theta}) \\ {\bf y}({\bf x}, {\bf u}, t; {\boldsymbol{\theta}}) = {\bf g}({\bf x}, {\bf u}, t; {\boldsymbol{\theta}}) \, . \end{gather}\] Suppose we have an infinite, error-free record of data, and a correctly specified model structure (composed of parametric relationships between variables). Is it possible to obtain a unique estimate for each parameter? If not, and if alternative parameter estimates produce very different predictions, we cannot make predictions with confidence. This may cause our study to be an unproductive use of time, effort, and resources.

I have considered the problem for model structures used in various biological settings.

Analysis of linear switching systems representing (e.g.) biosensor data:

Optical biosensor experiments1 provide a means of indirectly observing the interactions of biomolecular species in real time. The experimental setup features an immobilised analyte bound to a sensor surface, and an analyte in solution made to flow over this surface. A series of “kinetic” experiments aim to determine the rate constants of interactions. Experiments typically consist of two or more phases, delineated by a change in experimental conditions. For example, in the association phase, some concentration of analyte is made to flow over the surface for a specified time. In the dissociation phase, the solution is changed to buffer (zero analyte concentration).

In certain cases, the experimental output is appropriately modelled by a linear switching system (LSS). A LSS is a collection of linear time-invariant state-space systems, with a switch that determines which system is in effect at each time point. A schematic of the LSS output is shown in Figure 2 below.

A schematic of optical biosensor response for an experiment of tho phases, with the change in injected analyte concentration shown.

Figure 2: A schematic of optical biosensor response for an experiment of tho phases, with the change in injected analyte concentration shown.

As standard methods in identifiability testing and parameter estimation are not appropriate for an LSS structure, it is necessary to design other methods.

See the abstract of my PhD thesis:2
“Global a priori identifiability of models of flow-cell optical biosensor experiments”, Bulletin of the Australian Mathematical Society 98, no. 2 (2018): 350-352.


Recent publications on identifiability analysis

Recent publications on aspects of disease modelling

Code contributions


R Shiny Apps

Talks and posters on identifiability analysis

Approaches to analysing jump dynamical systems in biomolecular kinetics (and elsewhere?)“, talk presented at the ANZIAM Mathematical Biology Special Interest Group (MBSIG) meeting, February 14th, 2022.

“Structural identifiability analysis for switching system structures: towards a toolkit for changing times”, talk presented at Dynamical Systems Applied to Biology and Natural Sciences (DSABNS 2022 Virtual), February 10th, 2022. (DSABNS Contributed Talk Award.)

Hot off the virtual press...

Figure 3: Hot off the virtual press…

“Numerical investigation of structural minimality for structures of uncontrolled linear switching systems with Maple”, talk presented at the Maple Conference 2021 (virtual), November 2021.

“My Enemy, My Ally: how useful is this mathematical model?”, talk presented at the ARC Centre of Excellence for Mathematical & Statistical Frontiers (ACEMS) Early-Career Researcher Retreat, November 3rd, 2020.
Talk begins with a brief comparison of mathematics and science fiction.

“Branching out into structural identifiability analysis with Maple”, Maple Conference 2020 (online), Nov 2nd 2020.

“Frustrated mathematical modelling and changeable destinies: Structural identifiability analysis of models to support useful results”, Seminario de Investigación Interdisciplinar para la Innovación en Ciencia y Tecnología, (SICTE Interdisciplinary research, Catholic University of the North, Chile), invited oral presentation online, Oct 20th 2020.

“An introduction to the testing of model structures for global a priori identifiability (with examples drawn from Plasmodium falciparum malaria modelling)”, invited oral presentation for Influencing Public Health Policy with Data-informed Mathematical Models of Infectious Diseases, Creswick, Victoria, July 1st 2019.

“Biological modelling, and rarely asked questions of the 21st century”, poster presentation, BioInfoSummer, University of Western Australia, Perth, December 3rd 2018.

“Biological modelling, and rarely asked questions of the 21st century”, poster presentation, Australian Bioinformatics and Computational Biology Society Conference, Melbourne, November 26th 2018.

Work in progress

Whyte, J. M. “Structural minimality of linear swtiching system structres, as motivated by flow-cell optical biosensors and biomolecular interactions”

  1. For a cool video introduction, look here↩︎

  2. Won’t you?↩︎