PhD Studentship: Next-generation machine learning for genetic interaction prediction with lab-in the-loop validation in pancreatic cancer

University of Exeter; Exeter; On-site; GBP 20,000 - 24,000; Full time

Benefits offered by this job

Funded PhD studentship

Job summary

The University of Exeter invites applications for a fully funded PhD studentship in computational genomics focusing on genetic interactions in cancer. The project combines machine learning with perturb-seq data to build prediction models and test them in pancreatic cancer contexts, in collaboration with the FGC and AstraZeneca/Cancer Research Horizons.

The successful candidate will work on scalable architectures, HPC workflows, and rigorous evaluation protocols, contributing to open-source

Qualifications

• Strong background in machine learning applied to genomics.
• Experience with CRISPR/genomics data analysis.
• Experience with HPC environments and large datasets.

Responsibilities

• Develop GI prediction models using perturb-seq data.
• Evaluate graph neural networks, autoencoders, and other architectures for joint gene embeddings.
• Assess scalability to genome-wide prediction and generalisation across cancer types.
• Collaborate in a lab-in-the-loop cycle with wet-lab partners for pancreatic cancer data.

Skills

Genomics data analysis

Machine learning

CRISPR genomics

Single-cell sequencing

High performance computing

Education

PhD in genomics / bioinformatics

Tools

Python

R

PyTorch

TensorFlow

Jupyter

Job description

Background

Cancer cells often depend on specific genes to survive, making those genes attractive drug targets. CRISPR screens knock out genes to measure their importance for cancer cell survival, and large efforts such as DepMap and Project Score have mapped individual single‑gene dependencies across many cancer models. However, single‑gene studies miss vulnerabilities that only appear when two genes are disrupted together. These genetic interactions (GIs), including synthetic lethality, are a rich but under‑explored source of oncology targets.

As testing ~400 million gene pairs is infeasible, computational prediction is essential. Current GI prediction methods combine limited pairwise screens with larger single‑gene screens but rely on low‑dimensional readouts and generalise poorly across disease contexts(1).

By contrast, a proof‑of‑concept study showed that perturb‑seq, which couples CRISPR editing with single‑cell transcriptomics, can support GI prediction using recommender‑system approaches(2). Near genome‑wide perturb‑seq datasets are now available(3, 4), providing transcriptome‑wide readouts that could enable scalable, accurate prediction models.

Objectives

1. Build next‑generation GI prediction models using large‑scale perturb‑seq data. Initial model development will use publicly available perturb‑seq datasets, including large‑scale data from a leukaemia cell line and induced pluripotent stem cells(3, 4). You will evaluate graph neural networks, autoencoders, and other architectures capable of learning joint gene embeddings from perturb‑seq profiles. You will assess scalability to genome‑wide prediction, handling of sparsity and batch effects, and generalisation across cancer types.
2. Insights into genetic interaction in pancreatic cancer. Pancreatic cancer is one of the deadliest cancers, with average survival at diagnosis under six months. The AstraZeneca and Cancer Research Horizons Functional Genomics Centre (FGC) will generate perturb‑seq data and GI screens in pancreatic cancer models. These experiments will support a "lab in‑the-loop" cycle, where new measurements fine‑tune the computational model in the pancreatic cancer context. Note that researchers at the FGC will contribute the experimental data and "wet‑lab" work, while the student project is purely computational.

Expected outcomes

• A validated computational framework for GI prediction, released as open source software.
• Experimentally validated high confidence genetic interactions in a pancreatic cancer cell line.
• Insight into how GI predictions transfer across biological contexts.
• A reproducible and rigorous workflow integrating computation and experiment in a “lab‑in‑the‑loop” cycle, while controlling bias and false discovery rates.

You will gain

• Expertise with modern machine learning methods applied to genomics.
• Expertise in single cell and CRISPR genomics.
• High performance computing skills.
• Industry aligned experience through FGC involvement, including exposure to data quality standards, model evaluation, and target prioritisation.

An interdisciplinary project

The main supervisor is an expert in statistical and machine‑learning methods for genome editing; the co‑supervisor specialises in statistics, reproducible coding and model evaluation. FGC supervisors (Dr Kalinka, Dr Ross‑Thriepland, Dr Walter) will provide additional expertise in genome editing, genomics, cancer biology and bioinformatics.

Facilities and Resources

The University of Exeter will provide high performance computing for model development and training. The FGC will provide world‑leading CRISPR and single cell screening facilities and technical expertise.

Funded by: MRC Medical Research Council

£21,805. The studentship will provide funding of fees and a stipend of £21,805 per annum for 2026/7.