Current Projects
The following projects are being undertaken between April and August 2025.
Finding Healthy Online: What the Eyes Reveal
In partnership with Ocado Retail Ltd.
Online food retail platforms are becoming primary options for consumers’ daily dietary choices. Harnessing such platforms to improve the quality and sustainability of food choices can enable digital food environments that encourage healthier and more sustainable options.
For this project, the interdisciplinary team are using data science and eye-tracking data to describe real-life consumer’s decision-making and website use during online supermarket food shopping.
It builds on the ‘Finding Healthy Online’ study which collected eye-tracking data between February and August 2024 from ten shoppers whilst they purchased groceries using the Ocado website at Manchester Metropolitan University. We aim to understand the shoppers’ website navigation patterns, their website interactions, and engagement with nutrition and sustainability information. By integrating insights from nutrition and data science, this project explores the influence of website design and digital food environment on consumer behaviour in partnership with Ocado Retail.
The project uses eye-tracking datawhich offers rich and detailed information about the customer’s real-world website use, with fine temporal granularity, including gaze times (fixations) and eye-movement directions (saccades) over specific areas of interest (webpages, pictures, prices, nutrition information, etc.). Analysing this data with purchase receipt information, pre-shop survey data and transcripts of shopping journey interviews, we explore “how” people use the supermarket website for food shopping.
The outputs will support ongoing research engagement with Ocado and inform their health strategy.
By Gauri Venkatachalapathi
Identifying Genetic Predictors of Vascular Complications in Giant Cell Arteritis
In partnership with National Institute for Health and Care Research (NIHR) Leeds Biomedical Research Centre (BRC).
Giant Cell Arteritis (GCA) is the most common type of vasculitis affecting older people. It occurs when the arteries (particularly those at the side of the head) become inflamed leading to a narrowing of the vessels and reduced blood supply. The main symptoms include headaches, fever, scalp tenderness, and jaw pain when eating or talking. In more severe cases, and if not treated promptly, it can lead to serious complications like vision loss or stroke (ischaemic complications). Early diagnosis is essential; however, GCA can be difficult to diagnose with the initial symptoms often resembling those of other conditions.
The first line of treatment for GCA patients is with glucocorticoids (steroid medication). High-dose therapy is started on diagnosis, with even higher doses recommended for those with ischaemic complications. On subsequent tapering of the medication, relapse is seen in about half of all patients with GCA. This treatment is essential but there is also a considerable risk of complications arising due to long-term glucocorticoid treatment, with 86% of patients experiencing glucocorticoid toxicity in the form of cardiovascular or metabolic disease.
By identifying those at the highest risk of serious side effects, we could offer them additional treatments, such as tocilizumab. These treatments are currently considered too expensive for general use but may be cost-effective for high-risk patients before they have a major relapse. Some studies have shown that age and cardiovascular risk factors are linked to an increased risk of ischaemic complications in GCA. Building on this work, the aim of this study is to identify genetic predictors of GCA and recognise individuals at the highest risk. The study will use data from the UKGCA consortium, including patient records and genome-wide genotypic data.
By Lynette Linzbuoy
Scalable Visualization and Explainability of Synthetic Datasets
In partnership with 4-Xtra Technologies Ltd, a University of Leeds start-up.
This project is part of a larger “Making visualization scalable for explaining machine learning classification models” (MAVIS) project that is funded by the EPSRC (EP/X029689/1).
Driven by the current lack of available visualization techniques that are efficient at scale for various types of data, this DSDP project will explore and implement new methods for visual comparison and evaluation of real versus synthetic data.
Specifically, the project seeks to improve the visual assessment of data quality and model performance, including the discovery of hidden patterns, detection of anomalies or outliers, and more generally, evaluation and validation of faithfulness of AI-generated synthetic data in relation to realworld counterparts.
Using a combination of publicly available datasets and synthetic data generated by 4-Xtra’s proprietary models, the project will develop and validate fast, scalable visual tools to enable intuitive exploration of data similarity and dissimilarity.
The project fosters interdisciplinary collaboration and has potential applications across sectors such as business analytics, healthcare, and environmental science. By improving the clarity and accuracy of data comparisons, it can inform better decision-making and risk assessment. In the longer term, the outcomes are expected to increase trust in synthetic data and support policy development through more reliable and explainable AI.
In summary, this project is positioned to make significant contributions to the field of generative AI by advancing our ability to visually compare and evaluate real versus synthetic data, thereby enhancing the fidelity, utility, and explainability of generative models.
By Netochukwu Onyiaji
Psychiatric History and Routinely Assessed Outcomes of Hospitalisation (PHARAOH)
In partnership with NHS West Yorkshire Integrated Care Board (ICB) -Leeds
About 30% of general hospital inpatients have both mental and physical health conditions (Mental-Physical Multimorbidity-MPM). Compared to patients with only a physical illness, MPM patients experience worse outcomes, like longer hospital stays, more complications, and increased emergency readmissions. This leads to poor patient experiences, increased strain on limited hospital resources and contributes to greater health inequalities, particularly among older individuals and those from deprived communities.
Therefore, there is a pressing need to understand how MPM leads to poorer outcomes across different patient groups. This insight will aid in the design and implementation of targeted, cost-effective interventions that improve patient care and reduce NHS costs.
Data on patients’ physical and mental health are often separated. This proof-of-principle project will use pseudonymised data from the Leeds Data Model (LDM) linking primary and secondary health and social care data to understand patients’ trajectories and subsequent hospital outcomes. We will use an A to B to A longitudinal approach to map patients’ journeys:
- A: Identify a cohort of patients with a recent mental health condition in primary care (registered with Leeds GPs)
- B: Track these individuals through secondary care (LTHT hospital admissions).
- A: Re-identify these patients in primary care post-discharge, including outcomes such as length of stay and 30-day readmission.
The primary focus of this work is to:
- work collaboratively with the NHS West Yorkshire ICB to actively identify primary care patients with mental health disorders.
- establish the feasibility of tracking patient pathways through secondary care using the Leeds Data Model.
By Precious-Gift Alele
Can we estimate rock strength from rock texture alone?
In partnership with the University of Leeds School of Earth & Environment, British Geological Survey, Geosolutions Leeds.
As the Energy Transition accelerates, the demand for useable geo-mechanical data – especially the strengths of different potential reservoirs, aquifers, or storage sites – will massively increase. Traditional methods for obtaining this data are time-consuming and expensive. Developing faster, cheaper, and data-driven approaches will support decarbonisation and the transition towards net zero.
We are developing a state-of-the-art data mining tool to extract rock mechanics data (grain size, porosity, and strength) from a large corpus of published papers and open access theses. Using machine learning and statistical analysis, we will test the hypothesis that rock strength can be predicted from grain size and porosity alone.
Objectives:
- Build a state-of-the-art data mining tool to extract useful rock mechanics data from published papers and open theses.
- Use a statistically significant dataset to test the hypothesis that rock strength can be predicted using only the grain size and porosity.
The results of this systematic data science analysis will be of interest to the wider geological and engineering communities. The analysis is highly likely to provide sufficient pilot data for a follow-up proposal for UKRI funding (EPSRC and/or NERC) and will support organisations engaged in subsurface geological and engineering activity for the Energy Transition.
Sadiq Balogun