MRC DTP in Precision Medicine

Decoding the Tumour Microenvironment from H&E Slides using Spatially-Aware AI

Supervisors

Prof John Le Quesne, School of Cancer Sciences, University of Glasgow
Dr Ke Yuan, School of Cancer Sciences, University of Glasgow
Prof David Chang, School of Cancer Sciences, University of Glasgow

Industry Partner: TileBio

Summary

Deep Learning AI for Spatial Biomarker Discovery in Cancer
The complex cellular architecture of tumours dictates clinical outcomes, yet this information is not fully leveraged from standard diagnostic slides. This project aims to bridge this gap by developing an AI model that can infer detailed spatial-omics data directly from clinical routine Haematoxylin and Eosin (H&E) images.

The core of the project is the development of a transformer-based model that treats tissue architecture as a language. By learning the contextual relationships between cells—the ""grammar"" of tissue organisation—the model will move beyond single-cell analysis to understand the entire tumour ecosystem. The model's development will utilise very large collections of multimodal pathology data for initial self-supervised pre-training, followed by fine-tuning for the specialised task of multi-omics prediction from H&E.The primary outcome will be a validated ""Virtual Multi-Omics"" platform for analysing large clinical cohorts. This platform will be applied to discover and validate novel spatial biomarkers, such as immunosuppressive niches or unique morphological signatures, that predict patient outcomes and response to immunotherapy. This work is situated at the interface of machine learning, cancer biology and therapeutic development, offering advanced training in a highly translational environment with strong industrial connections.