Dr Peter Wijeratne
Probabilistic modelling of neurodevelopment and degeneration
Understanding and predicting neurodegenerative disease progression using machine learning
Disease progression can be viewed in terms of a sequence of events that characterise the transition of observable features from a healthy to an unhealthy state. Recently, statistical models and machine learning – collectively termed “disease progression models” – have emerged as powerful tools to understand and predict disease progression using data from patients with dementias, such as Alzheimer’s disease (AD) and Huntington’s disease (HD). In particular, disease progression models provide a natural framework to explore and characterise the relationships between progression and external variables, such as risk and protective factors, which could potentially inform monitoring and treatment planning.
In this inter-disciplinary PhD project the student will use a state-of-the-art disease progression model, the Temporal Event-Based Model [1], and apply it to datasets of AD and HD to extract new information about disease progression and its associated risk and protective factors. In both diseases the student will have access to several datasets with multiple biomarkers, including multi-modal magnetic resonance imaging (structural, diffusion weighted, functional); biofluids (blood plasma, cerebrospinal fluid); cognitive, psychiatric and behavioural test scores; and genetics (GWAS, somatic instability). In addition to application of the model, there is also scope to improve its capabilities, for example by incorporating risk and protective factors directly into the model as causal factors.
Mapping trajectories of early life neurodevelopment using machine learning
Early life neurodevelopment is key to downstream function later in life, such as cognition and motor ability, which impacts on an individual’s ability to lead a healthy and long life. However, little is currently known about the timing of changes in key features of neurodevelopment – such as myelination in the brain as measured by magnetic resonance imaging (MRI) – and how this timing influences downstream executive function. Machine learning presents an opportunity to map trajectories of early life and explore the interaction between these developmental trajectories and associated risk and protective factors.
In this inter-disciplinary PhD project the student will use a state-of-the-art progression model, the Temporal Event-Based Model [1], and apply it to datasets of neurodevelopment to extract new information about developmental progression and its associated risk and protective factors. The student will have access to several datasets with multiple biomarkers, including multi-modal magnetic resonance imaging (structural, diffusion weighted); electroencephalograms; microbiome; cognitive, psychiatric and behavioural test scores; and rich demographic characterisation. In addition to application of the model, there is also scope to improve its capabilities, for example by incorporating risk and protective factors directly into the model as causal factors.
General
The projects are computationally-focused and would suit a student with reasonable mathematical and programming skills, and a keen interest in machine learning and modelling neurodevelopment or degeneration. To support the interdisciplinary nature of the project, and in particular students from a less technical background, training will be provided in computational modelling, data analysis, and statistics. An appropriate co-supervisor will be appointed based on the specific project collaborators and the student’s research interests. As well as collaborating within Sussex Neuroscience, students will also benefit from being part of the Predictive Analytics Lab, and will have interactions with clinical groups at University College London and University of Capetown.
Key references
- [1] Wijeratne, PA et al. The temporal event-based model: Learning event timelines in progressive diseases. Imaging Neuroscience (2023): 1-19. DOI: 10.1162/imag_a_00010
Visit Peter's profile page for a full list of publications.