Back to news

Exploring Novel Applications of AI for Research and Innovation – announcing our 2024 funded projects

09 December 2024

We are delighted to announce the 13 projects funded through this years’ Accelerate Science and Cambridge Centre for Data Driven Discovery (C2D3) funding call.

This year’s call took place over the summer and received over 100 applications from teams across the University. Successfully deploying AI to tackle real-world challenges requires effective interdisciplinary collaboration, supported by time and resources to bring together potential research partners, develop new AI tools and software toolkits, and develop new skills or networks. Recognising that this work often falls outside the scope of routine funding calls, the Accelerate and C2D3 funding programme aims to help to fill this gap by offering small grants that can be deployed flexibly.

Selected teams will all start or scale interdisciplinary projects that contribute to advancing the use of data science and AI for research and innovation. The wide range of projects supported through this call will catalyse further discoveries to accelerate scientific progress and create AI tools that are capable of delivering benefits for science and society. Projects range from developing tools to automate and optimise scientific workflows to applied research exploring the impact of LLMs on student learning and the use of machine learning to reduce illegal hunting in sub-Saharan Africa. Details of all 13 projects can be found below.

Projects will be delivered over the next year and we will share updates though the Accelerate Programme blog. Accelerate Science and C2D3 will offer technical support and facilitation as part of project delivery.

Translating emerging findings to impact

To translate promising approaches or prototypes developed by projects funded through the Accelerate-C2D3 funding call into real-world impact, in 2024 we have also trialled an Impact funding scheme. With the support of this scheme:

• Dr Anne Alexander (Cambridge Digital Humanities) and collaborators at the Pulitzer Centre will work with the Rainforest Investigations Network, non-profit investigative journalism organisation Watershed Investigations, and local journalists from Uganda and Northern Ireland to build capacity in the use of machine learning for public interest journalism.
• Dr Edward Harding (Institute of Metabolic Science) will develop a digital platform to increase the accessibility of research into pre-clinical phenotyping and drug discovery.
• Dr Ines Prata Machado and Eleanor Wolmark (Department of Oncology) will convene patients academics and policy makers to develop resources that communicate the current state of AI research in medical sciences to the public and UK policy makers.
• Professor Felix Steffek (Faculty of Law) and his team will engage legal professionals with their technical work using AI to predict the outcome of court cases.

Projects awarded in 2024:

Harnessing AI for metagenomic discovery of cryptic pathogen lineages

Alexandre Almeida, MRC Career Development Fellow, Department of Veterinary Medicine

The human intestinal tract is colonised by a community of microorganisms — the human gut microbiome — with beneficial roles to human health. However, many microbial species inhabiting the human gut have the potential to cause disease including species implicated in severe infections and antibiotic resistance worldwide. This project aims to leverage high-resolution metagenomic strategies powered by AI to pre-emptively identify and track emerging pathogen variants.

Automating Data Analysis with Large Language Model Agents

Boris Bolliet, Assistant Teaching Professor in the MPhil in Data Intensive Science, Department of Physics, Cavendish Astrophysics Group

Many researchers work in large scientific collaborations where they must process, analyse, and interpret vast volumes of data using advanced statistical methods, including Machine Learning and AI-based techniques. The project team will explore whether it is possible to use Large Language Model (LLMs) to automate and optimise scientific workflows. The goal of this project is to develop cmbagent, a Multi-Agent System powered by pre-trained LLMs and designed to automate complex data analysis tasks.

Multi-modal Foundation Models for the Early Detection of Neurodegenerative Diseases

Zhongying Deng, Post-doctoral Research Associate at the Department of Applied Mathematics and Theoretical Physics

Neurodegenerative diseases like Alzheimer’s and Parkinson’s are chronic, incurable conditions that gradually impair brain function. Current diagnostic methods often rely on a single type of data, limiting their ability to detect only a single neurodegenerative disease. This project addresses this limitation by using multi-modal data, e.g., MRI, PET, and genomics data, to train large-scale foundation models to improve early detection.

Enhancing the deterrence effect of anti-poaching patrols in protected areas using machine learning approaches

Charles A. Emogor, Schmidt Science Postdoctoral Fellow, Departments of Zoology and Computer Science and Technology

Anil Madhavapeddy, Professor of Planetary Computing Department of Computer Science and Technology, University of Cambridge

This project aims to apply machine learning methods to help reduce illegal hunting plaguing thousands of protected areas globally. Using existing data spanning decades from seven protected areas in three sub-Saharan African countries, the team will build prototype models to help park rangers identify where and when to patrol. This work spans savanna and forest habitats and will be the first multi-site effort using spatial data from rangers, allowing the team to capture the socio-economic variability in the distribution of poaching related threats.

Unlocking Educational Potential: Exploring the Impact of LLM Chatbot Tutors on Student Learning Psychology and Behaviours

Megan Ennion, PhD Student, Faculty of Education

Ros McLellan, Associate Professor, Faculty of Education

Despite a growing body of research suggesting LLM tutors hold considerable potential to enhance student learning experiences and outcomes, important gaps remain due to the pace of technological advances. This study focuses on the effect of LLM tutors on student learning psychology and key behaviours and offers a preliminary exploration of the methodological potential of LLMs in educational research.

Predictive Modelling of OCD Subtypes and Comorbidities to Enhance Personalized Treatment Strategies

Máiréad Healy, PhD Student, Department of Psychology

Zoe Kourtzi, Professor of Experimental Psychology, Department of Psychology

OCD is a complex and heterogeneous condition, where comorbidities such as Autism Spectrum Disorder and Attention-Deficit/Hyperactivity Disorder complicate diagnosis and treatment, leading to suboptimal outcomes in clinical settings. This project aims to advance the diagnosis and treatment of OCD by developing a sophisticated machine learning model to accurately identify subtypes and associated comorbidities enabling personalized treatment strategies to improve patient care and outcomes.

Reconstructing Patient Journey for Early Detection of Endometriosis-Associated Ovarian Cancer

Dr Golnar Mahani, Research Associate, Department of Oncology, Cancer Research UK, Cambridge Institute and Early Cancer Institute

Ovarian cancer is the most lethal gynaecologic malignancy, with less than half of patients surviving five years post-diagnosis. This poor prognosis reflects lack of early detection strategies as well as rapid emergence of chemoresistance. This project aims to develop new early detection strategies by focusing on endometriosis-associated ovarian cancer (EAOC). By reconstructing patient journeys using large language models the project aims to identify potential EAOC cohorts and gain critical insights into early signs of malignancy.

AI-powered Demand Forecasting for Enhanced Healthcare Resource Allocation in the NHS

Feryal Erhun, Professor of Operations and Technology Management, Judge Business School

Zidong Liu, PhD Student, Judge Business School

This proof-of-concept project explores the potential of AI in predicting and managing healthcare demand across the UK’s primary and secondary care sectors. The NHS grapples with escalating service demand due to an ageing population and resource constraints, exacerbated by the COVID-19 pandemic. This project addresses these challenges by developing predictive models that accurately forecast healthcare demand, enabling proactive resource allocation and improved patient outcomes.

Exploring Interdisciplinary Frontiers: Cognitive Science, Computational Modeling, and Artificial Intelligence

Runhao Lu, PhD student (Gates Cambridge Scholar), MRC Cognition and Brain Sciences Unit

Alexandra Woolgar, Programme Leader, MRC Cognition and Brain Sciences Unit

This project aims to foster academic exchange and collaboration among leading scholars and early career researchers in the UK and EU, within the fields of AI, cognitive neuroscience, psychology, philosophy, computer science, and robotics. By facilitating partnerships between institutions, the team aim to advance both AI’s role in understanding brain function and how neuroscience can inspire new AI frameworks. Through interdisciplinary collaboration, the project seeks to bridge the gap between these rapidly evolving fields, driving innovative solutions and expanding the impact of cognitive science and related domains.

Voice in the Machine: Harnessing Speech AI for Naturalistic Prosody in Audiological Assessment

Alexis Deighton MacIntyre, Research Fellow (Leverhulme Trust), MRC Cognition and Brian Sciences Unit

Lidea Shahidi, Research Associate, MRC Cognition and Brain Sciences Unit

Auditory disabilities affect people from all walks of life, and the World Health Organisation estimates that nearly 2.5 billion people will live with some degree of hearing loss by 2050. Clinicians and researchers rely on listening tests to assess speech perception. Breakdown between tests and realistic communicative scenarios may misalign patient expectations with outcomes and contribute to under-diagnosis of hearing disorders. This project will explore the experimental control and generative capabilities of state-of-the-art speech synthesis to produce audiological testing materials and conduct evaluation to ensure fit for use in clinical tests.

AI meets cultural heritage – Non-invasive imaging and machine learning techniques for the reconstruction of degraded historical sheet music

Anna Breger, Senior Postdoctoral Researcher, Department of Applied Mathematics and Theoretical Physics

Carola-Bibiane Schönlieb, Professor of Applied Mathematics, Head of Cambridge Image Analysis group, Department of Applied Mathematics and Theoretical Physics

This project will explore the possibilities, challenges and limitations of imaging and machine learning methods for reconstructing degraded historical sheet music. Such degradations may happen due to chemical or physical damage. The project team will form a novel collaboration network spanning libraries, imaging laboratories and AI imaging researchers. Samples will be selected of musical manuscripts of historical interest and the team will employ advanced imaging systems to scan these manuscripts and apply standard as well as newly developed, advanced machine learning methods to reconstruct the degraded parts.

Mitigating Confounding Variables in MRE Brain Imaging through Domain-Invariant Contrastive Learning

Jakob Träuble, PhD Student, Department of Chemical Engineering and Biotechnology

Gabriele Kaminski Schierle, Professor in Molecular Biotechnology, Department of Chemical Engineering and Biotechnology

Magnetic Resonance Elastography (MRE) is a promising brain imaging technique that measures the mechanical properties of brain tissue and shows potential for detecting neurological changes earlier than traditional MRI. However, its broader application is hindered by significant variability in datasets that distort machine learning models trained on such data. This project tackles this issue by developing an ML framework to account for and reduce the influence of these confounding variables with the goal to establish a more reliable framework for analysing MRE data, potentially improving the early diagnosis and treatment of neurological diseases.

Online training of large-scale Fortran-based hybrid computational science models, with applications in climate science

Joe Wallwork, Research Software Engineer, Institute of Computing for Climate Science (ICCS), University Information Services

Jack Atkinson, Senior Research Software Engineer, Institute of Computing for Climate Science (ICCS)

Many fields make use of large-scale Fortran codes that have been developed over decades through international research efforts, for example numerical weather and climate prediction. Recent advances in machine learning have opened up several exciting opportunities to advance these models. However, this presents a significant software challenge as much machine learning is conducted in Python. The team has developed a software solution - FTorch - that allows Python-based ML models written in PyTorch to be easily interfaced with Fortran codebases. This project seeks to extend the functionality of this software to facilitate online training of models.