The abstract is read at the publisher; the summary is JClub's.
CASTLE improves spatial domain boundary detection in spatially resolved transcriptomics by using cell-type-aware contrastive embeddings, achieving competitive or superior accuracy and stability across tissues and resolutions.
- Why it matters: Spatial clustering is central to spatial transcriptomics, but delineating spatial domain boundaries remains difficult, limiting precise tissue microenvironment interpretation.
- What they did: CASTLE builds a spatial graph from spatial proximity, then refines it using cell-type similarity or expression similarity when cell types are unavailable, and trains with a local-context contrastive objective.
- The result: Across multiple tissues and technology resolutions, CASTLE generally matches or exceeds prior methods on multiple datasets and metrics, enabling more accurate, consistent domain detection and finer-grained downstream subtype differentiation.
The findingWhy it mattersWhat they didThe result
AI summary of the abstract · may contain errors · not medical advice · read the paper.