spSheaf: sheaf-theoretic spatial domain inference from raw molecules without cell segmentation

spSheaf, a segmentation-free sheaf-theoretic method for MERFISH hypothalamus, achieves mean ARI 0.571 and NMI 0.655, ranking first of twelve on NMI.

• Why it matters: Spatial-transcriptomics often needs cell segmentation and discards 20–60% of transcripts, motivating methods that infer spatial domains directly from raw molecules.
• What they did: spSheaf models local expression subspaces and uses alignment-adjusted disagreement to attenuate transitions during clustering, then evaluates on MERFISH and additional annotated sections.
• The result: It leads SSAM, Points2Regions, and FICTURE by 0.16 ARI on held-out slices; across three further sections it leads on one, ties on one, and trails SSAM by 0.06 ARI, with sheaf weighting adding +0.102 ARI for distinct programs.

AI summary of the abstract · may contain errors · not medical advice · read the paper.