In October 2024 we ran a funding call offering rapid access to short-term pump priming funding to University of Bristol researchers pursuing innovative applications of AI in health or biomedical research.
The hope is that the funded activities will lead on to developing and submitting external bids for future research programmes and projects that use or address AI in health and biomedical research contexts.
Here we announce the successful applicants from this call – find out more about the latest AI in Health awardees and their projects…
Asme Boussahel – Cryptic Chatter: Decoding Multicellular Interactions with AI Microscopy
Obesity affects 13% of the global population and is linked to 2.8 million deaths annually. Despite extensive research, the mechanisms driving obesity, especially the cellular interactions in adipose tissue remain unclear.
Asme explains: “This knowledge gap is partly due to the overreliance on animal models, which often fail to translate to human outcomes. Advances in 3D cell culture and bioprinting enable the recreation of complex cellular environments in-vitro, providing a powerful tool for understanding tissue-specific mechanisms.
“I have developed a novel human in-vitro adipose tissue model by co-culturing adipocytes and macrophages in an extracellular matrix (ECM)-mimicking hydrogel. This model replicates changes in cellular function and cell-cell and cell-ECM interactions in obesity. However, characterising these interactions remains challenging without isolating the cells from the 3D tissue-mimicking environment, which risks altering their phenotype and limiting the insights gained.
“I have adopted state-of the art live imaging methods, which allow real-time observation of cellular dynamics in 3D over extended periods. Analysing such models is inherently complex due to the presence of ECM components, multiple cell types, and the long culture durations. The use of imaging dyes is often unreliable, failing to distinguish between cell types. Moreover, the imaging generates massive datasets, making manual analysis inefficient and limiting.
“To address this, I am employing AI and machine learning algorithms to analyse these complex imaging datasets. AI models enable segmentation of cells, classification, and tracking of interactions in both healthy and obesity-induced conditions, providing deeper insights into the cellular mechanisms that contribute to obesity and metabolic dysfunction.”
Jeff Clark – Quantifying explainable AI interpretability for healthcare settings
AI is increasingly deployed to support decision-making in healthcare settings. Explainable AI (xAI) plays a critical role in helping users to understand AI suggestions. However, most xAI research is targeted at AI developers and technical practitioners, with limited focus on how healthcare professionals engage with xAI systems.
Jeff explains: “This gap is significant, as explanations must be quickly and intuitively understood, given the time constraints and cognitive demands placed on healthcare professionals. Without effective explanations, healthcare workers are unlikely to trust AI systems, limiting their potential to improve patient care. Safe AI implementation in healthcare requires systems that meet clinicians needs, including xAI components that clearly explain processes and outputs.
“This project builds on recent work conducted with clinicians at Bristol Royal Infirmary (BRI), where requirements for xAI systems were gathered [1]. We aim to quantify how well clinicians understand different types of explanations, measuring their speed and accuracy in interpreting them. These findings will advance the design of xAI systems, fostering trustworthiness and supporting integration and adoption in clinical workflows, enhancing the impact of ongoing work designing clinical dashboards with AI components. The work will generate new insights into xAI usability, enhance clinician-AI interaction, and improve patient care.”
[1] Clark, Jeffrey, et al. “Exploring the Requirements of Clinicians for Explainable AI Decision Support Systems in Intensive Care.” arXiv preprint (2024).
James Hodge- Feasibility of Artificial Intelligence (AI) for Patient Registries
Patient registries are fundamental to healthcare research and clinical practice, enabling the collection, analysis, and sharing of patient data to study outcomes, assess treatment efficacy, and inform policy. However, the rapid expansion of healthcare data presents significant challenges, including the manual effort required for data curation, integration of diverse data sources, and maintenance of data quality.
James explains: “Artificial Intelligence (AI), and more specifically, natural language processing (NLP), has the potential to transform the development and management of patient registries by automating processes and enabling real-time insights. This research aims to evaluate the feasibility of AI in addressing these challenges and optimising patient registry workflows.
“Currently, we are hosting a national registry for TSC (Tuberous Sclerosis Complex) at the University of Bristol. TSC is a multisystem, chronic health condition which is associated with tumour growth in various areas of the body leading to significant physical and mental health complications. The purpose of the national registry is to document the manifestations and progression of this condition across paediatric and adult patients. Data entry will be performed manually, as usual, across 20 TSC specialist clinics in the UK.
“This project will fund a research fellow to conduct a feasibility study. The aim is to establish a pilot study to assess the feasibility of using AI in the TSC patient registry. The key research questions are: 1. How effectively can AI automate the extraction and integration of structured and unstructured data from multiple healthcare sources?; and 2. Can AI-driven quality assurance methods improve data consistency and completeness in patient registries?”
David Murphy – AI Prediction Of Peptide Ligand-surfacome Interactions – Proof Of Concept
Artificial intelligence promises to reveal many biological secrets based on its ability to predict the complex interactions between proteins that govern cell function. Particularly important in this regard is the way the cells communicate with each other over long distances using protein hormones that circulate in the blood that interact with distant target receptors located on the surface of cells that mediate biological responses. These long-distance biological signals can go wrong in disease.
David explains: “The collection of human cell surface proteins is called the “surfaceome” and consists of ~4000 proteins, most of which are not understood in terms of their functional interactions. Conversely, there are lots of proteins circulating in the blood that have no known function, but presumably interact with a binding partner protein to elicit a physiological effect. One such “orphan ligand” is the 39 amino acid glycopeptide copeptin. Whilst circulating copeptin has emerged as a biomarker for many chronic cardiovascular conditions and associated metabolic dysfunctions, its normal physiological functions are not understood, and its role in disease remain a mystery.
“We will carry out two complementary screens to identify surfaceome copeptin binding partners. The first will be computational, the second will be in cell culture. We will partner with Isambard-AI, one of the most powerful computers in the world, to carry out a deep learning-based in-silico pulldown screen to model putative copeptin protein-complex interactions using state-of-the-art protein interaction prediction methods; and identify putative copeptin receptors using an unbiased cell-based proteo-genomic CRISPR activation screen.”
Jess Wheeler -Using generative AI (language models) in qualitative health research: Building collaborations and developing a framework to securely test and evaluate the use of generative AI in qualitative health research, and impacts on people who are part of minoritised, marginalised communities and most impacted by health inequalities?
Generative AI language models (e.g. GPT or BERT) are rapidly emerging global technologies, already in use in qualitative health research (QHR), promising increased efficiency and breadth of data inclusion, in research that is notoriously time and labour intensive. However, frameworks for ethical oversight, quality evaluation and standards of practice in the use of language models in QHR, are hardly yet established.
Jess explains: “A major concern is the use and impact of language models in QHR, on the evidence-base and associated healthcare advances, relevant to the health and lives of diverse, minoritised, marginalised populations, already subject to health inequalities. Currently, use of language models in QHR is piecemeal, with researchers making use of available tools despite numerous potential limitations, including: limited technological performance; hidden and inaccessible methodologies and algorithms; and proprietorial ownership issues. These limitations make most platforms entirely inappropriate for testing and evaluating outputs in relation to confidential, highly sensitive, QHR data.
“Through a series of collaborative meetings, we will draw on local expertise (in language model architectures, qualitative health research, healthcare ethics, and in patient and public inclusion and involvement) to critically evaluate ethical, technical and practical concerns, and to create a framework for building a safe, high-performance, responsive, generative AI platform, with appropriate ethical oversight, suitable for testing and evaluating language model QHR outputs, specifically in relation to those most impacted by health inequalities.
“This collaborative development project is the first stage towards seeking funding to build a secure world-class AI platform, able to securely incorporate extensive QHR datasets.”
Upcoming AI in Health workshops:
Artificial intelligence in Health: socio-digital transformation, ethics and governance
Tuesday 6 May 2025 11:30 – 13:30, Life Sciences Building, Tyndall Avenue, BS8 1TQ
More info and book
Artificial intelligence in Health: genomics, protein design and drug discovery
Thursday, 22 May 2025 11:30 – 13:30, Life Sciences Building, Tyndall Avenue, BS8 1TQ
More info and book
Keep up to date with events and opportunitites – Join our AI in Health research community