About the project
An aging population incurs huge socioeconomic cost due to neurodegenerative diseases, for several of which there is no treatment. This project will address the challenge of interpreting molecular, genetic, and imaging data using machine learning models.
Of particular interest is in quantifying uncertainty in model-based predictions.
Vast amounts of data relating to genetic, molecular, imaging and spectroscopy methods relating to neurodegenerative diseases is becoming available, posing the challenge of early diagnosis and developing a better understanding of their origin and spread along neural connections.
Techniques from machine learning are being increasingly applied to problems in inference from biological and medical data. A particular issue in the use of machine learning models is quantifying the uncertainty associated with model-based predictions. The need is at two levels:
- (a) given an individual patient, how certain can al algorithm be in the predictions it makes;
- (b) at a population level, what can a designer of an algorithm guarantee about its performance.
The former is addressed using techniques based in Bayesian statistics and the latter in the framework of conformal predictions. Mechanistic modelling capturing how misfolded proteins diffuse and data-driven modelling from machine learning will be used, and uncertainty quantification approaches developed.
This project will explore these techniques, further develop algorithms suitable for their application to neurodegenerative diseases, where progression of the pathology takes place over a graph, specified by neuronal networks in the brain.
The project will use data from public repositories as well as local data collaborating with clinicians at the University Hospital Southampton, in consultation with clinical expertise.