Mechanistic mathematical models are powerful tools for describing complex biological systems and can significantly aid in improving the understanding of their underlying intricate (biochemical) interactions, predicting their dynamic responses and optimising their performance. Increasing model complexity, either in terms of structure or number of kinetic parameters adopted, commonly leads to unnecessary model structure complexity, poor identifiability, and/or over-parameterization. Identifiability problems, related to both structural and practical identifiability have been studied over the past decades, especially in relation to systems biology models.Nevertheless, a re-parametrisation of models that are found to be structurally non-identifiable, that retains meaningful biological traits, is still an open problem, especially for realistic large-scale models. This PhD project will comprise research on the identifiability of models of biochemical systems. As mentioned above, such models are typically over-parametrized, containing a much larger number of parameters than the number of measured (or measurable state variables) making estimation of meaningful parameter values hard and/or misleading. The project will develop new methodologies for improving model identifiability, leading to more robust and reliable system predictions.
We will focus on both structural and practical identifiability and in particular the development of efficient methodologies for re-parametrisation of non-identifiable models, which, as mentioned above, is an open problem, especially for large-scale, stiff nonlinear models. Although model reduction is mentioned in the literature [e.g. ref] this entails mainly ad hoc model/parameter simplifications. We will exploit the most efficient exiting algorithms for structural (global and local) identifiability (e.g. StructuralIdentifiability and STRIKE-GOLDD) and we envisage employing low-order projections of the non-identifiable models, exploiting fast dynamics and/or dissipativity properties to develop simplified re-parametrised systems, while fully maintaining their essential physical traits, i.e creating physically meaningful connections between the original and the reduced model(s). For practical identifiability analysis the project will exploit the latest approaches and tools as well as profile likelihood and Monte Carlo sampling, focusing on machine-learning methods for fast and intelligent data reconciliation in the case of sparse and/or noisy data, as well as intelligent optimisation methods to investigate and alleviate the interdependence(s) of large groups/classes of parameters.
The project is a collaboration between the University of Manchester and the University of Mons in Belgium and in particular the group of Prof. Alain Vande Wouwer (https://staff.umons.ac.be/alain.vandewouwer/activitesEn.html) where the successful candidate will spend at least a year of the programme.
The successful candidate should have an excellent academic backgound evidenced by Distinction-level (or equivalent) undergraduate and postgraduate degrees in Chemical Engineering or a relevant discipline, as well as experience in mathematical modelling and computational programming. Evidence of research experience in terms of research projects, publications, conference presentations, is also desirable.