PhD Studentship: Developing machine learning models for radiation dose prediction using third-generation sequencing data

Award details

The project will focus on developing the next generation of models used for radiation exposure dose prediction, using machine learning methods to extract information from “long-read” sequencing data produced in UKHSA labs.

3-year PhD studentship in the UK Health Security Agency under supervision of Dr Philip Davies, and King’s College London Department of Informatics under supervision of Dr Sophia Tsoka.

The studentship is funded by the NIHR Health Protection Research Unit in Radiation Threats and Hazards. The Health Protection Research Units are NIHR’s flagship research partnerships between Universities and UKHSA, focused on the highest priority challenges in public health. From 1st April 2025, the HPRU in Radiation Threats and Hazards, led by Imperial College London, will seek to advance understanding of ionising radiation and health and delivery direct impact on policy to improve the long-term health of the nation.

Recent advances in DNA and RNA sequencing technology, in particular within third-generation sequencing platforms (ONT), have led to a huge amount of biological data being easily produced from a small sample of blood or skin. Specifically, the cellular state of an individual can be assessed by measuring gene expression levels, detecting DNA mutations, identifying RNA modifications, and uncovering RNA splicing perturbations, all from a single sample. This unprecedented breadth of information promises to revolutionise personalised medicine, with huge implications for radiation protection.

The aim of this PhD project is to develop novel computational methods which integrate different types of data. With support from the Department for Informatics (KCL) and the Radiation Effects Department (UKHSA), the successful candidate will develop new models using our continuously expanding datasets obtained from laboratory experiments and real-world samples of individuals exposed to radiation. These models will ultimately contribute to improved triage during emergencies.

The project will focus on developing the next generation of models used for radiation exposure dose prediction, by leveraging cutting-edge machine learning methods to extract information from “long-read” sequencing data produced in UKHSA labs. While traditional models use measurements of a handful of previously identified indicator genes, our data contain measurements on thousands of genes, isoforms, mutations and modifications, whose co-variance could prove a powerful predictor of radiation exposure. Other variables such as time elapsed since exposure to radiation, gender, age or infection status, could also be inferred from the data and used as co-variates, thus further refining the model.

The successful candidate will assess each of the different types of measurement (gene expression, isoform usage, RNA modification and DNA mutations) for their utility as measurements of radiation exposure, both as independent predictors as well as jointly. As these data types span thousands of features, (e.g., 30k genes, 120k isoforms, thousands of DNA mutations and RNA modifications), it will be important to identify the features that confer highest predictive power. A variety of Data Science methodologies will be used to explore the structure of the data and identify clusters with similar patterns of behaviour. State-of-the-art predictive methods such as informed machine learning will also be employed to test whether advanced techniques outperform classical models in terms of accuracy, speed and precision. Using prior knowledge from public databases may inform ML models in order to constrain the parameter space and evaluate predictions using classical dose-calibration models as a benchmark. These results will be directly used to guide the design of additional laboratory experiments, which will be conducted by scientists at UKHSA labs in parallel to this project and will in turn generate additional data to feed into and improve models.

Award value

Stipend: £22976 per annum

Bench Fees: £1,000 per annum

Tuition fees: £7,963 per annum

Other: travel expenses to conferences

Eligibility criteria

Applicants should hold, or achieve by the start of the programme, an undergraduate degree or MSc in Computer Science or Bioinformatics, with a UK First- or Upper Second-Class honour grade. In exceptional cases, other qualifications and experience may be considered and all applications will be assessed on their merit as appropriate to the individual case.

This studentship is only available to candidates who are eligible for home tuition fee status. Applicants from overseas will be considered if they are able to  cover the studentship costs by either self-funding, other fellowship, and/or through Government funding.

Application process

Deadline of 31^st March 2025 for applications, successful candidate to start their studentship no later than 1^st October 2025

To be considered for the position candidates must apply via King’s Apply online application system. Details are available here.

To apply for this studentship funding please cite the code ‘IDUKHSATsoka2501’ in the Funding section of the application form. Please select option 5 ‘I am applying for a funding award or scholarship administered by King’s College London’ and type the code into the ‘Award Scheme Code or Name’ box. Please copy and paste the code exactly.

Applicants should email Dr Philip Davies and Dr Sophia Tsoka before submitting the application, and include a summary of their academic background, research interests and any relevant previous experience.

Contact Details

Philip R. Davies, Philip.R.Davies@ukhsa.gov.uk

Sophia Tsoka, sophia.tsoka@kcl.ac.uk