Research Experience Scheme for UG Students

Duration: 8 weeks (part-time available)

Preferred start date: 15 June 2026

Project team: Andrew McIntosh andrew.mcintosh@ed.ac.uk (project supervisor)

Location: Institute of Neuroscience and Cardiovascular Research, Chancellors Building, Royal Infirmary of Edinburgh

Project summary

The student would be developing a tool to search literature databases (e.g. PubMed, Google Scholar) to identify studies with samples and data related to mental health ‘omics, in order to add new studies to an existing table and to annotate this for beyond-genomics ‘omics for those studies already listed. The existing table can be found here: https://datamind.org.uk/datamind-data/discoverable-data/genetics-table/ 

We have a list of studies that have been identified separately by searching adncontacting our collaborators, against which we could test the accuracy and sensitivity of any pilot tool developed as part of this project.

Ideally, we are looking to develop a piece of software that could complete updates on a fixed frequency basis rather than as a one-off, although a one-off piece of work would also be beneficial.

Project objectives

1. Provide a simple tool that can identify molecular phenotyping (omics) studies that include mental health information
2. Test the ability of the tool to identify studies we already know exist from our previous work
3. Identify likely false-hits, studies that are identified by the tool, which do not include relevant data

Learning outcomes

The intern will gain experience of working to a project brief, collaborating with other colleagues at other institutions and working with large language models

Candidate requirements

Essential criteria:

1. A superficial familiarity with LLMs such as ChatGPT
2. At least a superficial experience of coding and the need for careful version control
3. The ability to work collaboratively as part of a team

Desirable criteria:

1. Experience of using GitHub or other version control softward
2. Experience of working with LLMs
3. An interest in mental health research

Duration: 4 weeks (part-time available)

Preferred start date: 31 July 2026

Project team: Steven Kerr steven.kerr@ed.ac.uk (projec supervisor), Bryan Lopez De Munain (PhD tutor)

Location: Centre for Medical Informatics at Usher Institute, Edinburgh Bioquarter

Project summary

Advances in artificial intelligence, particularly deep learning and transformer-based architectures, are rapidly transforming the field of medical image analysis. Retinal imaging offers a powerful, non-invasive window into human health, capable of revealing signs of ophthalmic diseases such as diabetic retinopathy, glaucoma, and age-related macular degeneration, as well as broader systemic conditions. The growing availability of large retinal image datasets presents a valuable opportunity to develop and evaluate machine learning approaches capable of supporting earlier diagnosis and improved clinical decision-making.

This internship is aligned with ongoing research at the University of Edinburgh in partnership with the Scottish Collaborative Optometry–Ophthalmology Network eResearch (SCONe). This initiative is building a large-scale repository of retinal images collected through community optometry practices across Scotland to enable innovation in eye health research and early disease detection. The resource is designed to support the development of computational tools that can assist clinicians in identifying disease at earlier stages and improve population health outcomes.

Vision Transformers can achieve strong performance on a variety of image recognition tasks. These models are increasingly being explored in medical imaging due to their ability to capture global contextual information within images and to scale effectively with large datasets. However, implementing and optimising such models for retinal imaging applications remains an active area of research.

Project objectives

The aim of this internship project is to implement and evaluate transformer-based computer vision models for the analysis of retinal images.

The project will involve fine-tuning an open-source transformer model for retinal image classification and assessing its ability to detect conditions such as diabetic retinopathy and other retinal pathologies. This will be done using openly available repositories of retinal images. The intern will also explore methodological variations that may influence model performance and computational efficiency, such as ultra-wide-field retinal images with conventional colour fundus photographs, image resolutions, and experimenting with masking or tokenisation strategies that may can reduce computational cost.

Learning outcomes

The internship will provide the student with hands-on experience in applying modern machine learning techniques to real-world biomedical imaging problems. By the end of the project, the intern will have gained experience in implementing transformer-based architectures for computer vision tasks, managing and preprocessing medical image datasets, and designing computational experiments to evaluate model performance. The intern will also develop practical skills in deep learning frameworks commonly used in AI research, as well as an understanding of the methodological considerations involved in deploying machine learning models in healthcare contexts. This includes issues such as dataset variability, model interpretability, and computational efficiency.

Candidate requirements

Essential criteria: Some experience coding in Python

Duration: 8 weeks (part-time available)

Preferred start date: 15 June 2026

Project team: Rik Sarkar rsarkar@inf.ed.ac.uk, Chris Wood chris.wood@ed.ac.uk (project supervisors), Handing Wang (PhD Tutor)

Location: School of Informatics, Informatics Forum.

Project summary

Drug molecules are designed to bind with specific target proteins, but they might also bind to other proteins known as off-targets. This is an important challenge in drug design, as off-target binding cause dangerous side effects.

The student will categorise and tabulate datasets and models commonly used in drug design, affinity and off target effect prediction. They will summarize the attributes of various datasets and models and where they are applicable, and create visualisations (e.g. Hierarchies, PCA and clustering) where appropriate. They will identify compatible dataset-model pairs.

The student will measure the efficiency of various models. They will estimate their running time and training (or fine tuning) time for models.

Project objectives

- Map the standard datasets of drug molecules, drug affinity prediction, off target effects, and relevant proteins. Tabulate their attributes. Create visualisations.
- Map the common models in drug design and off target prediction similarly.
- Identify datasets that are compatible with important models
- Estimate the resource requirements (e.g. time) to run and train (or fine tune) the various models. Note that the Estimate does not necessarily require a full training cycle, and can be estimated from typical epoch times.
- Produce a report that can be used by researchers to identify datasets and models suitable to their research problem.

Learning outcomes: experience in surveying a research field and resources, model training and evaluation, data visualisation, writing.

Candidate requirements

Essential criteria: some knowledge of machine learning/data science and programming.

Desirable criteria: experience in working with neural networks.