AI and Cancer PhD: Predicting Pancreatic-Cancer Gene Interactions with Lab-in-the-Loop Validation

AI and Cancer PhD: Predicting Pancreatic-Cancer Gene Interactions with Lab-in-the-Loop Validation

Quick Decision

A strong computational project, if the validation design is clear. The student will develop models to predict gene-pair interactions and test them against pancreatic-cancer data produced by an experimental partner. The advertised methods include graph neural networks, autoencoders and joint gene embeddings; evaluation covers genome-wide scale, sparse data, batch effects, transfer across cancer types and false-discovery control. The student’s role is computational, while partner researchers carry out the wet-lab work. The project links model development to experimental cancer research and provides access to Exeter high-performance computing and an industry-linked functional-genomics centre. Official project description

The deadline needs written clarification. The project record gives 8 December 2026, while the official 2027 award page states 5 January 2027; the project PDF gives no separate date. Until Exeter confirms which applies, work to the earlier date. On international fees, the current national FGx-DLA postgraduate page explicitly says international students at the University of Exeter do not need to supplement UKRI funding to cover international fees. Check the exact 2027/28 project fee and that the final offer reflects this current policy. Award page · National FGx-DLA postgraduate page · PGR fee schedule

Research Problem and Doctoral Work

What Is Unsolved and Why It Matters

Cancer cells can depend on particular genes for survival. A single-gene screen can identify one dependency at a time, but it can miss vulnerabilities that appear only when two genes are disrupted together. These non-additive genetic interactions include synthetic lethality: a pair of perturbations may impair survival even when either perturbation alone has a smaller effect. Testing combinations across the genome creates a very large search space, so computational ranking is useful if it can generalize to new genes and disease contexts.

The project proposes to use Perturb-seq, which combines CRISPR perturbation with single-cell RNA sequencing. Rather than reducing each perturbation to one growth score, it captures transcriptome-wide profiles for individual cells. A 2019 proof of concept used rich single-cell phenotypes for selected gene interactions and recommender-system modelling to organize and predict interactions. Genome-scale resources now provide much larger single-gene perturbation maps, including a leukemia-cell dataset and an iPSC resource. These make the project timely, but they do not by themselves solve the central modelling question: a single-gene map is not an exhaustive pairwise interaction truth set. Norman et al., Science · Replogle et al., Cell · Feng et al., Cell Genomics

Work, Methods, and Potential Outputs

The advertised work has two connected stages:

1. Develop and compare prediction models. Begin with public Perturb-seq data, including a leukemia cell line and induced pluripotent stem cells. Evaluate graph neural networks, autoencoders and other architectures that learn joint gene embeddings from perturbation profiles. Assess whether the models scale toward genome-wide prediction, handle sparse measurements and batch effects, and transfer across cancer types.
2. Use new pancreatic-cancer measurements in an iterative modelling cycle. AstraZeneca and Cancer Research Horizons' Functional Genomics Centre will generate Perturb-seq data and genetic-interaction screens in pancreatic-cancer models. These measurements will be used to fine-tune and validate computational predictions in the pancreatic context. Intended outputs include an open-source prediction framework, experimentally validated high-confidence interactions in a pancreatic-cancer cell line, analyses of context transfer, and a reproducible workflow that controls bias and false discovery. These are expected outcomes, not guaranteed results. Official project description

The division of work is clear: the student’s PhD is computational. FGC researchers supply the experimental data and wet-lab work; Exeter supplies high-performance computing. Prospective students should therefore want to build models and analyze data, while also being comfortable working closely with experimental collaborators. The project description does not specify how many model-development cycles are planned, how candidate pairs will be selected for each screen, or how many pairs and replicates the pancreatic screen will contain.

A central design question is where the models will obtain pairwise interaction labels. The cited genome-scale Perturb-seq resources primarily map single-gene perturbations, but the model needs effects measured or inferred for gene pairs. Norman et al. provide an earlier pairwise Perturb-seq proof of concept, and newer dual-guide or multiplex CRISPR systems can scale pair screening, but this project's exact training labels, GI score definitions and held-out test sets are not named. Norman et al. · Dual-guide screen study · in4mer multiplex platform

School, Programme and Field Context

The official project document identifies the Department of Mathematics and Statistics and lists the University of Exeter as the primary institution. Some summary information associates the opportunity with Clinical and Biomedical Sciences, so applicants should confirm the formal administrative school and registration route. Department research themes include Mathematics for Health and Life Sciences and Statistics & Data Science, with links to interdisciplinary institutes and PhD activity. The project is part of the multi-institutional MRC Functional Genomics Doctoral Landscape Award, which offers shared training, scientific meetings and collaboration across six UK institutions. Exeter is the project's stated primary institution; no joint or dual award is stated. Project description · Department research · MRC programme

Exeter's published overall positions are 136th in the QS World University Rankings 2027 and 157th in the Times Higher Education World University Rankings 2027. These measure the whole university, not this project's subject. More relevant institutional evidence is Exeter's REF 2021 profile: Biological Sciences had 41% of submitted research rated world-leading (4*) and 49% internationally excellent (3*); Mathematical Sciences had 42% 4* and 52% 3*; Computer Science and Informatics had 33% 4* and 50% 3*. The department reports that 100% of its mathematical-sciences research impact was rated internationally excellent in the 2021 exercise. REF assesses institutional submissions, not this supervisor's individual group or the day-to-day experience on this project. Exeter rankings · Exeter REF 2021 results

The UK labour-market context is favorable for hybrid computational-biology skills, although it is not a promise of a particular job. Skills England's 2026 life-sciences assessment projects 66,000 additional jobs in priority occupations between 2025 and 2035, and about 118,000 total demand including replacement, while also warning that AI-related forecasts are uncertain. The combination of genomics, statistical modelling, software reproducibility and model evaluation could transfer to computational biology, bioinformatics, functional-genomics data science, or biotech/pharma R&D. Sector estimates are national, not specific to this PhD or its graduates; no placement or employment outcome is guaranteed. Skills England life-sciences assessment

Supervisor and Lab Assessment

Supervisor Research Directions

The named lead supervisor is Dr Magdalena Strauss, a senior academic in Mathematics and Statistics. Her institutional profile describes research at the intersection of mathematics, data science and biomedical science, including statistical modelling of heterogeneous cellular responses to stress, disease and drugs and how those responses depend on mutations. This aligns with the project's need to distinguish biological response from variation due to cell state, guide efficacy, donor or batch. Dr Strauss's profile · Academic staff directory

Recent Important Publications

• Feng et al. (2026), “A genome-scale single-cell CRISPRi map of trans gene regulation across human pluripotent stem cell lines.” The study builds a large CRISPRi single-cell map across genetic backgrounds and measures how guide identity, knockdown efficacy, donor and cell line contribute to observed responses. It is directly relevant to the iPSC data and shows why cross-context models need careful adjustment. It is a single-gene map rather than pairwise GI ground truth, and the authors note that more donors are needed for reliable attribution. Full text
• Strauss et al. (2025), “A single-cell and tissue-scale analysis suite resolves Mixl1's role in heart development.” The authors introduce COSICC statistical tools for single-cell perturbation analysis in complex populations and use knockout chimeras to study developmental lineages. This demonstrates the lead supervisor's work on statistical analysis of heterogeneous perturbation responses. It concerns developmental knockout effects, not cancer pairwise interactions or the proposed deep-learning models. PubMed
• Cooper, Coelho, Strauss et al. (2024), “scSNV-seq: high-throughput phenotyping of single nucleotide variants by coupled single-cell genotyping and transcriptomics.” This work connects a cell's actual edited genotype with its transcriptomic phenotype, addressing the risk of treating guide presence as proof that the intended edit occurred. It is relevant to genotype assignment and single-cell data quality, but focuses on single-nucleotide variants rather than gene-pair interactions. Genome Biology
• Coelho, Strauss et al. (2024), “Base editing screens define the genetic landscape of cancer drug resistance mechanisms.” The study uses functional screens and transcriptomic analysis to map cancer mutations associated with drug-response phenotypes. It shows recent cancer-focused functional-genomics experience in the wider research network, while its unit of analysis is variant-by-drug response rather than pairwise gene knockout. Nature Genetics

These papers support the team's strength in statistical perturbation analysis, single-cell methods and cancer functional genomics. They do not establish that the proposed large-scale pairwise prediction problem has already been solved by this group; the PhD's central method contribution remains an open research task.

Lab, Resources, and Collaboration Environment

The official project document names Dr Kalinka, Dr Ross-Thriepland and Dr Walter from the Functional Genomics Centre as contributors with expertise in genome editing, genomics, cancer biology and bioinformatics. The centre is described as a collaboration between AstraZeneca and Cancer Research Horizons, with CRISPR and single-cell screening capabilities. Exeter will provide high-performance computing for model training. The project also describes exposure to data-quality standards, model evaluation and target prioritisation. The specific co-supervisor's name, division of supervisory responsibilities and expected visit or placement schedule are not provided. Project description · Functional Genomics Centre

Funding, Application, and Living Cost

The MRC FGx-DLA page advertises up to 18 funded PhD places across six institutions for September 2027 entry. Applications opened on 26 October 2026. The current official award page lists Tuesday, 5 January 2027 as the deadline, while the project entry records 8 December 2026; the project-specific PDF has no separate date. Ask the lead supervisor and Exeter PGR admissions which date applies, and keep 8 December as the working deadline until you have a written answer. The programme permits up to three project preferences in the questionnaire but only one application per person. Official award page · MRC programme page

The award page states four years of funding, with part-time study also available. It lists UK tuition fees, a UK Research Council minimum stipend of £21,805 per year for 2026/27 (updated annually), and up to £5,000 per year for research training and support. That extra allowance is research/programme support, not a personal living stipend. Exeter's 2026/27 research-fee page lists UK fees of £5,238 and international science fees ranging from £23,800 to £29,900, depending on topic. The current national FGx-DLA postgraduate page explicitly says Exeter-based international students do not need to top up UKRI funding to cover international tuition. The published UK-to-international fee spread is therefore a rate reference, not an expected applicant payment under the current programme policy. Confirm the exact 2027/28 project fee and that the final offer reflects the current policy, as annual rates and award terms can change. UKRI stipend and fee rates · Exeter research fees · National FGx-DLA postgraduate page

For living costs, Exeter estimates £1,300–£2,000 per month for one student in Exeter or Cornwall, based on local spending in April 2026. The current stipend averages about £1,817 per month when divided by twelve. This leaves a simple estimated monthly balance of about £517 at the low-cost endpoint and a shortfall of about £183 at the high-cost endpoint. Actual housing and lifestyle costs vary, and that comparison excludes family expenses, visa/relocation costs, travel, deposits, emergencies and future inflation. Exeter living-cost estimate

The published entry requirements are a minimum 2:1 honours degree or international equivalent in a relevant subject. Accepted areas include genetics, molecular biology, biomedical science, human biology, biochemistry, medicine/healthcare, bioinformatics, computer science and statistics. The usual English threshold is IELTS 7.0 overall with no component below 6.5 or equivalent, and evidence is due by the application deadline. The portal application requires transcripts, a maximum two-page CV, two references and the FGx-DLA applicant questionnaire. Applicants can list up to three project preferences but must submit only one application. If referee letters cannot be uploaded initially, provide referee details and check the later reference deadline; referees must not be from the prospective supervisory team. Application materials must be in English, with certified translated qualification documents where needed. Award entry requirements

Applicant Fit and Preparation

Required Background and Skills

The formal requirement is a relevant 2:1 degree or equivalent; the programme explicitly welcomes quantitative/computing and biomedical/genetics backgrounds. This is a fit for candidates who want to work primarily with code, models and experimental data. The official project names graph neural networks, autoencoders, single-cell and CRISPR genomics, high-performance computing, batch-effect handling, transfer across cancer types and bias/FDR control. It does not list a required programming language or a minimum prior wet-lab experience; the student project is computational.

Preferred Experience and Suggested Preparation

No project-specific preferred experience is formally stated. Useful preparation would include a reproducible Python or R analysis, a clear model comparison against simple baselines, and a concrete example of working with high-dimensional or noisy data. Candidates from computing or statistics can strengthen their application with basic cancer genetics and CRISPR-screen literacy. Candidates from biology can make their quantitative preparation visible through code, data analysis or statistical modelling. Be prepared to explain how to distinguish an interaction from two independent gene effects, how to avoid train/test leakage across genes or contexts, and how to control false discoveries. These are preparation suggestions, not additional eligibility rules.

Next Actions and Questions to Verify

Before Applying

1. Ask Dr Strauss and Exeter PGR admissions whether the project deadline is 8 December 2026 or 5 January 2027, and request the answer in writing. The project PDF lists Dr Strauss at m.strauss@exeter.ac.uk; the official award page gives pgrapplicants@exeter.ac.uk for application-process questions.
2. Ask for the exact 2027/28 project fee and confirm that the final award/offer applies the current FGx-DLA policy that Exeter international students do not need to supplement UKRI funding for international fees.
3. Ask which data or scores supply pairwise interaction labels for initial training and evaluation. The named large Perturb-seq resources are mostly single-gene maps.
4. Confirm the formal registration department, administrative school and Exeter campus; the project PDF names Mathematics and Statistics, while summary information also associates it with Clinical and Biomedical Sciences.
5. Confirm the co-supervisor's identity and how partner-centre scientists participate in supervision.

At Interview

• Which pancreatic-cancer model(s), pair-screen design, number of pairs, replicate depth and single-cell readouts are planned?
• How will the work define a genetic interaction and separate it from single-gene effects?
• What gene-held-out and cancer-context-held-out tests will be used? Which simple baselines, negative-control strategies, batch corrections, calibration and FDR thresholds are expected?
• How many lab-in-the-loop rounds are planned, and who chooses the next experimental pairs?
• What coding and single-cell methods are taught during the programme, and what computational support is available?

Before Accepting an Offer

• Confirm in the written award letter the annual stipend, tuition amount, international-fee responsibility, duration and annual increases.
• Confirm the registration institution, required partner-site visits or placements, and whether travel is funded.
• Check the final start date and whether the project will remain four years full-time or use an agreed part-time schedule.

Evidence and Sources

Project, Programme and Research Environment

• Official project description
• MRC FGx-DLA award page and entry requirements
• National FGx-DLA postgraduate page and international fee policy
• MRC programme page and project list
• Mathematics and Statistics research themes
• Academic staff directory
• Exeter Research Software and Analytics Group
• Exeter institutional rankings
• Exeter REF 2021 results

Supervisor, Papers and Experimental Context

• Dr Magdalena Strauss profile
• Functional Genomics Centre
• Feng et al., genome-scale iPSC CRISPRi map
• Strauss et al., COSICC single-cell perturbation analysis
• Cooper et al., scSNV-seq
• Coelho et al., cancer drug-resistance screens
• Pancreatic cancer statistics

Methods and Field Outlook

• Norman et al., Perturb-seq genetic-interaction manifolds
• Replogle et al., genome-scale Perturb-seq
• Synthetic-lethality model benchmark
• Systema perturbation-prediction evaluation framework
• Deep-learning models versus simple perturbation baselines
• Large dual-guide GI screen
• in4mer multiplex GI-screening platform

Funding, Cost and Labour Market

• UKRI stipend and fee levels
• Exeter postgraduate research fees
• Exeter living-cost estimates
• Skills England life-sciences workforce assessment