Early career fellowship 2026-27

Lead supervisor: Ben Faber

Topic description: Understanding the evolutionary drivers of pelvic shape

Project background:

Evolution through natural selection mirrors the framework of Mendelian randomisation, because alleles change in frequency as a consequence of the phenotypes that they influence in turn influencing fitness. However, when traits are phenotypically or genetically correlated it can be difficult to determine which features are under selection. Pelvic shape is complex, and is of particular evolutionary interest due to its critical role in maternal and neonatal survival but the drivers of its development are poorly understood. This project will use pelvic shape derived from ~70,000 UK Biobank participants, epidemiology and Mendelian randomisation to characterise the drivers of pelvic shape evolution.

Please contact Ben Faber in the first instance.

 

Lead supervisor: Yi Liu

Topic description: CAUSE-GRAPH: Causal Analysis Using Semantic Extraction and Graph Representation of Mendelian Randomization Studies

Project background:

Tens of thousands of Mendelian randomization (MR) studies have been published, yet their findings remain fragmented across the literature without systematic integration. We have demonstrated that large language models (LLMs) can efficiently extract structured MR findings from publication abstracts (MR-KG; https://doi.org/10.64898/2025.12.14.25342218). In this AI-focused project, the fellow will extend this approach to extract comprehensive MR evidence from full-text articles, integrate the data into a knowledge graph, and apply graph learning methods to identify potential novel causal relationships. The project will focus on one or more disease areas aligned with the applicant's research interests and those of the IEU.

Please contact Yi Liu in the first instance.

Lead supervisor: Alexandros Rammos

Topic description: Developing a sequencing analysis pipeline for Cleft Collective parent-offspring trios: A pilot study of structural variation and de novo variant annotation.

Project background: 

The Cleft Collective is the world’s largest cleft lip/palate research cohort yet lacks whole genome sequencing (WGS) data essential for identifying structural variation and de novo variants contributing to orofacial clefts. De novo mutations are implicated in many developmental conditions but remain understudied in clefts. Oxford Nanopore Technology (ONT) offers key advantages: detection of structural variants missed by short-read approaches, and simultaneous methylation calling from native DNA. Before pursuing cohort-wide ONT sequencing, we must develop robust analysis pipelines and establish feasibility across different sample types (blood vs saliva) and collection timepoints available in the Cleft Collective biobank.

Please contact Alexandros Rammos in the first instance.

 

Lead supervisor: Gemma Hammerton

Topic description: Vaping and smoking trajectories in young adulthood and early health impacts

Project background:

Electronic cigarettes are promoted as less harmful alternatives than smoking, however rapid uptake in youth and young adults, the quick evolution of products since their introduction in 2007, and the spread of misinformation has increased uncertainty around this. There is also limited evidence on near-term health consequences of vaping in young adulthood beyond cessation outcomes. It is therefore important to add to the body of evidence looking into differences in health outcomes between those with different nicotine use patterns. We will characterise nicotine-use trajectories and estimate associations with important health outcomes, while carefully accounting for prior smoking and key confounders.

Please contact Gemma Hammerton in the first instance.

Lead supervisor: Hannah Elliott

Topic description: Development of epigenetic prediction pipelines using DataSHIELD

Project background:

DNA methylation (DNAm), a stable but reversible chemical modification of DNA, is a promising biomarker of disease risk due to its capacity to change gene activity and to capture disease risk factors like smoking history and chronic inflammation. Unfortunately, the full potential of DNAm remains unexplored due to a focus on populations of European genetic ancestries and privacy/ethical restrictions that prevent data sharing between studies. This project adapts a federated analysis platform (DataSHIELD) for prediction work, collaborating with the Diverse Epigenetic Epidemiology Partnership (DEEP).  This work will advance privacy-preserving distributed epigenetic prediction work in the department across diverse datasets.

Please contact Hannah Elliott in the first instance.

Lead supervisor: Grace Power 

Topic description: Triangulating evidence on the effects of physical activity during pregnancy on maternal, pregnancy, and offspring health

Project background:

Physical activity during pregnancy is a complex exposure, and evidence of its causal effects on maternal and offspring health remains limited. Observational analyses are affected by residual confounding and have limited capacity to infer causality. RCTs are few and often small. Mendelian randomisation has rarely been applied to physical activity in pregnancy, though work to address this is underway. Target trial emulation has not been used in this context, and its implementation in this project is novel. Experimental work in animal models provides complementary mechanistic insight but is rarely considered alongside human analyses. This project will triangulate evidence across methodologies.

Please contact Grace Power in the first instance.

Lead supervisor: Gareth Griffith

Topic description: Quantifying bias due to confounding via geography across phenotypic and genotypic architecture in large-scale biobanks

Project background:

Clustering of common causes by geography can induce bias if ignored. Recent research has demonstrated that adjustment for genetic principal components (PCs)[1–3] incompletely removes bias from genetically predictions. However, the degree to which confounding via geography impacts genetically informed studies depends on the degree of clustering in exposure and outcome, often an unknown quantity.
Social and infection outcomes are particularly confounded and a systematic approach is sorely needed. We will generate a resource for researchers using methods developed by the supervisory team[4]; to demonstrate and quantify likelihood of bias due to confounding via geography for given relationships of interest.

Please contact Gareth Griffith in the first instance.

Lead supervisor: Fergus Hamilton

Topic description: Unlocking the genetics of infection history: Novel GWAS approaches for complex antibody data

Project background:

Antibody data are uniquely challenging because they capture two distinct biological processes: whether an infection occurred (exposure) and the magnitude of the immune response (severity). Standard GWAS fails here because antibody distributions are often non-normal or bimodal, conflating these two signals. We have recently developed 'funGWAS', a novel method utilizing quantile regression and mixture modelling to disentangle genetic effects on seropositivity from effects on antibody titre. Applying this to large-scale biobank data offers a way to identify genetic determinants of infection susceptibility and immune strength, which are currently indistinguishable in traditional analyses.

Please contact Fergus Hamilton in the first instance.

 

Lead supervisor: Kate Birnie

Topic description: Pre-pregnancy kidney conditions and adverse pregnancy outcomes: triangulating evidence from electronic health records

Project background:

Women with pre-existing chronic kidney disease (CKD) are at increased risk of adverse pregnancy outcomes (APOs) such as preeclampsia and preterm birth, yet evidence remains limited on how risks vary by CKD stage prior to pregnancy. Furthermore, little is known about other conditions, such as prior acute kidney injury, which may result in reduced kidney reserve, increased kidney vulnerability, and subsequent APOs. The Clinical Practice Research Datalink (CPRD) provides an opportunity to address these gaps for women entering pregnancy with prior kidney disease, explore heterogeneity in risk by ethnicity and deprivation, and assess whether associations are likely to be causal.

Please contact Kate Birnie in the first instance.