Summary
A bioRxiv preprint presents stEDGE, a computational framework that uses local tissue boundaries to reconstruct nested spatial domains and transition interfaces. The authors report results across 15 tissue sections and an application to cell-resolved profiling of fibrotic human lung.
A bioRxiv preprint introduces stEDGE, a computational method designed to reconstruct nested tissue domains and the transition interfaces between them from spatial transcriptomics data. The framework starts by estimating local boundary structure rather than treating a tissue as a collection of flat regions, then uses those boundaries to guide the identification of finer spatial domains.
The authors evaluated stEDGE across 15 benchmarked tissue sections covering different tissues, spatial-transcriptomic technologies and resolutions. They report that the method reconstructed developmental, inflammatory, tumour and brain architectures, and applied it to cell-resolved Xenium profiling of fibrotic human lung.
Why spatial boundaries matter
Spatial transcriptomics measures gene-expression patterns while retaining information about where those patterns occur within a tissue. This makes it possible to study tissue organisation rather than analysing cells or molecular profiles without their physical context.
A common computational approach is to divide a tissue into spatial domains, or regions with similar molecular profiles. The preprint argues that this flat representation can miss two important features: gradual transitions between regions and hierarchies in which smaller domains are nested within broader tissue compartments.
Those interfaces can be biologically meaningful. A boundary may mark a change between tissue states, while a transition region may contain cells or molecular programmes associated with remodelling, inflammation or disease progression. stEDGE is designed to represent stable compartments, boundaries, transition regions and localised niches within the same multiscale model.
How stEDGE builds a multiscale map
The method uses an edge-guided strategy. It first estimates local boundary structure in the spatial data and uses that information to guide reconstruction of fine-grained domains. This reverses the order used by domain-first approaches, in which regions are identified before their boundaries are explicitly considered.
stEDGE also introduces a domain transition index, or DTI, to quantify how strongly a domain is associated with transitions between spatial states. The authors combine DTI with inter-domain similarity and boundary strength to organise fine domains into a coherent hierarchy.
A tree-guided gene-attribution step then compares the molecular programmes associated with different levels of that hierarchy. It is intended to separate programmes shared across a parent domain from gene programmes that are specific to an individual branch or specialised spatial state.
Findings in fibrotic human lung
In cell-resolved Xenium data from fibrotic human lung, the authors report a multiscale hierarchy of lesion-associated remodelling. The reconstruction placed remodelling interfaces alongside a stable macrophage-associated airway or lumen state and spatially organised TLS-like lymphoid niches. TLS-like refers to tissue structures resembling tertiary lymphoid structures, where immune cells can form organised local assemblies.
The result illustrates the type of biological interpretation the framework is intended to support: a tissue lesion is represented not only as a changed region, but as a structured arrangement of persistent compartments, interfaces and specialised niches. The preprint presents this as a unified way to study how gene programmes and tissue organisation vary across spatial scales.
The work is currently reported as a bioRxiv preprint. Its findings and computational framework therefore remain at the pre-peer-review stage.