We are inviting applications for a fully funded 3.5-year PhD Computer Science studentship at the University of Warwick, jointly supported by GlaxoSmithKline (GSK), to work on an ambitious project at the intersection of machine learning, bioinformatics, and computational pathology.
Project Overview:
Integrating histopathological imaging with omics (e.g., transcriptomics, genomics, proteomics) holds tremendous promise in understanding disease mechanisms and improving clinical decision-making. Recent studies suggest that generative models can uncover latent structures and improve classifier robustness across modalities and populations.
This project (SpyGlass) will investigate machine learning methods, particularly generative and foundation models, to discover robust correlation structures in cross-modal datasets, especially within the cancer domain. The goal is to identify causally relevant links between tissue morphology and molecular profiles, potentially leading to new biomarkers or therapeutic targets.
Objectives:
- Framework Development: Design and implement a generative deep learning framework for cross-modal integration and analysis, resilient to distribution shifts.
- Correlation Discovery: Identify interpretable correlation structures across data modalities that relate to outcomes such as survival and treatment response.
- Validation & Benchmarking: Apply the framework across cancer types, evaluating various foundation models for generalizability and fairness.
Previous Work and Context:
This project builds on successful work under a previous GSK-Warwick studentship which linked WSIs to gene expression profiles in breast cancer using graph neural networks. Our current efforts extend this to additional cancers and modalities, such as multiplexed immunohistochemistry (mIHC), immunoflouresence, spatial transcriptomics and generative model-based domain translation, in collaboration with leading research institutions. This new studentship aims to develop the next generation of interpretable and cross-modal predictive models for cancer research incorporating large multimodal models.
Impact and Significance:
This research will contribute significantly to computational pathology and precision oncology by:
- Enabling robust, interpretable diagnostics across modalities
- Improving predictive accuracy in clinical decision support tools
- Contributing to equitable and generalizable AI for biomedical applications
Candidate Profile:
We are looking for highly motivated candidates with:
- A strong academic background in computer science, AI/ML, bioinformatics, or related fields such as mathematics and statistics
- Experience or strong interest in deep learning, generative models, or biomedical imaging/omics data
- Strong programming skills (Python, PyTorch, etc.)
- A commitment to interdisciplinary collaboration and impactful research
How to Apply:
If interested, please email Dr. Fayyaz Minhas (fayyaz.minhas@warwick.ac.uk) with the following:
- Your CV
- University transcripts
- A short statement explaining your interest in the project and how your background aligns with the objectives
Early expressions of interest are encouraged, as interviews may be held on a rolling basis. The candidates will need to satisfy all admission requirements for the University of Warwick.
Full fees and stipend at UKRI rates for 3.5 years plus a small budget for Travel and Conference attendance and equipment.
