Research
Two questions drive the lab. The first is what sets and holds a cell's identity, and whether that identity can be deliberately rewritten. The second is how cells instruct each other across immune, stromal and parenchymal boundaries. The four themes below are how we go after them: two on the biology, two on the methods that make it tractable. Disease and regeneration are where they meet.
Single-cell experiments now produce data at a scale where analysis is not a step after the experiment. It is a large part of the experiment. We build the frameworks that make that tractable: integrating datasets across studies, species and technologies, constructing reference atlases, and designing pipelines that stay reproducible from first exploratory run to final figure.
A recurring emphasis is honest handling of scale. Batch structure, ambient signal and uneven capture across cell types can each manufacture a convincing result that does not survive replication, so much of the methodological work is about knowing which differences are real.
Selected papers
Expression tells you what a cell is doing now; chromatin tells you what it is prepared to do. We use single-cell and single-nucleus RNA-seq together with ATAC and joint multiome assays to link the two, mapping accessible regulatory elements to the transcriptional programmes they control.
The practical goal is to identify the transcription factors that specify and maintain cell identity, and to work out which of them are merely correlated with a cell state and which are actually load-bearing. That distinction is what turns a descriptive atlas into a testable hypothesis.
Selected papers
Dissociated single-cell data tells you which cells are present but throws away where they were. Spatial transcriptomics and tissue imaging keep that context, at the cost of far harder analysis: segmentation, registration, sparse per-spot signal and measurements that only mean something relative to their neighbourhood.
We build machine learning methods for that setting: mapping cell state onto tissue architecture, detecting and quantifying vascular structure directly from images, and connecting spatial readouts back to the dissociated single-cell and epigenomic atlases that describe the same tissue.
We are wary of models that look good and change nothing. A clean embedding that suggests no experiment worth running has not helped. We use machine learning to generate hypotheses, and hold its output to the same standard of validation as everything else in the lab.
Selected papers
Endothelial cells are not passive plumbing. They are organ-specialised, transcriptionally distinct, and they instruct the cells around them. They supply niche signals that govern haematopoietic stem cell behaviour, tissue regeneration and immune recruitment.
We study how this vascular–immune dialogue is established, how it differs between organs, and how it breaks down. That includes inflammation, infection, ageing and the injured or regenerating tissue niche, where the endothelium is often both a driver of pathology and a route to repair.
Selected papers
Approach
We work on both questions in development, in stem cell and iPS-derived models, in inflammation, and in ageing and regeneration.
The approach is to measure broadly, then test. We combine high-dimensional and single-cell profiling across transcriptome, chromatin, protein and tissue space with functional assays in primary cells, organoids and tissue. Which assay we reach for follows from the question rather than from what the lab happens to be set up to run.
On the first question, we look for the regulators that set identity. What separates an endothelial cell from a hepatocyte, a tissue-resident macrophage from a freshly recruited monocyte, or the same nominal cell type in two different organs, is written by a tractable number of transcription factors. Finding them is the route to controlling identity rather than cataloguing it. We build cross-tissue atlases to generate candidates, use chromatin state to separate drivers from passengers, and then test the shortlist directly by perturbation. That is a reprogramming question and we treat it as one.
On the second, we study how cells instruct their neighbours. Endothelium governs which leukocytes adhere, transmigrate and are licensed once they arrive, and cytokine signalling sets the terms on both sides of that exchange. Parenchymal and stromal cells are not bystanders in it: the niche a macrophage or a stem cell sits in shapes what it becomes. We follow that dialogue across tissues and through ageing, where it turns inflammatory and tissue maintenance becomes tissue damage, which makes ageing the clearest setting for asking what regeneration would actually require.
Method development runs underneath all of it. We build the tools we need and release them, from one of the earliest pipelines for recovering paired full-length TCR sequences from single-cell data to deep-learning models that read disease state from tissue images more accurately than the clinical standard.
None of it works in isolation. Our closest collaborations are with experimental groups whose systems put real pressure on the analysis, and the traffic runs both ways: their samples sharpen our methods, and our predictions change what they do next.