Using artificial intelligence to extract retinal and neural information and to use Big Data multi-omics to discover biomarkers of disease.

(i) The retina is the only part of the body where blood vessels and nerves can be directly visualised. Measures of retinal arterioles predict cardiovascular disease, retinal venules are associated metabolic diseases such as diabetes, and the retinal nerve fibre layer thickness is associated with cognitive function. With the explosion of artificial intelligence in image analysis, this project aims to extract data from longitudinal retinal photographs and scans in over 6,000 participants from the TwinsUK cohort, taken over 20 years.
(ii) The purpose is to use novel deep learning techniques to create retina and neural datasets that would match existing omics information currently available in this unique and fascinating twin cohort, to enable “Big Data” modelling of common diseases in our populations.
The aims of the project are to extract longitudinal retinal measures and to examine available genomics, metabolomics, proteomics and microbiome datasets to look for biomarkers and predictors of diseases including diabetes, cognitive loss and cardiovascular disease. Associations will be validated using available datasets such as the UK Biobank study.
(iii) The student will develop a wide range of bioinformatics skills, including deep learning image analysis techniques (such as Automorph, already publicly available for retinal photographs), machine learning techniques such as random forest decision trees, and programming skills to use the many bioinformatics techniques needed for analysis of large, complex omics datasets.
(iv) For the first year the objective will be to complete a systematic review of retinal image analysis techniques, and to extract cross-sectional and longitudinal data on retinal blood vessels (including arteriolar and venous diameter, fractals and branching patterns) and neural measures (such as retinal nerve fibre layer thickness) and to perform simple linear regression studies on outcomes of interest such as cardiovascular disease and diabetes. The second year objectives will be to perform machine and statistical learning analyses of longitudinal metabolomic arrays matched to longitudinal retinal vessel measurements, to examine predictors of change and identify novel biomarkers across a wide range of phenotypes. The third year of the project will develop these analyses further, incorporating genomic and microbiome measures in the analyses to perform truly multi-omics studies with unique longitudinal datasets available within TwinsUK.