Research

Tumours contain molecular, cellular, spatial, and morphological information at very different levels. Our research asks how these different views relate to one another, and how we can best obtain the information needed to improve clinical decisions.

Spatial tumour biology

Tumours are complex tissues in which malignant cells coexist and interact with immune, stromal, and other cell populations. We use spatial and single-cell technologies to study tumour cell states, cellular neighbourhoods, and the spatial organisation of the tumour microenvironment, and to understand how these change during tumour evolution and treatment.

Our current work spans several cancer types, including pancreatic and ovarian cancer, neuroblastoma, and sarcoma, and uses technologies ranging from single-cell RNA sequencing to spatial transcriptomics and highly multiplexed imaging.

Computational pathology

Histopathology captures a remarkable amount of information about the underlying biology of a tumour. We develop machine-learning approaches that connect tissue morphology with molecular and cellular phenotypes, asking what can be learned about tumour biology from routine images.

We are particularly interested in combining histopathology with spatial molecular measurements to understand which aspects of tumour heterogeneity are visible in tissue morphology, and which require additional molecular information.

Interpretable machine learning

Machine-learning models can make increasingly accurate predictions from complex biological and medical data, but understanding what information they use to make those predictions remains challenging. We develop and apply interpretable machine-learning approaches to identify the biological and morphological concepts underlying model predictions.

Rather than treating interpretability only as a way to explain an individual prediction, we are also interested in using it as a tool for scientific discovery: to identify previously unrecognised patterns in tissue, relate learned representations to molecular states, and test whether models and human experts use the same information.