Statistical significance testing for spatial domain detection in spatial transcriptomics

SigDoTS controls type I error and improves spatial domain recovery in simulations, and in human dorsolateral prefrontal cortex it retains Layer 6 while merging Layer 3 candidates.

• Why it matters: Candidate domain counts can change with parameters, and testing on the same data used for discovery can create post-selection bias and anti-conservative p-values that promote oversegmentation.
• What they did: SigDoTS applies hierarchical hypothesis testing with family-wise error adjustment, modeling gene-gene covariance and spatial dependence, and computes p-values by rerunning the domain-detection pipeline on simulated data under a fitted single-domain null.
• The result: Under the single-domain null, it controls type I error at the root node and improves recovery when multiple domains exist; in dorsolateral prefrontal cortex it merges Layer 3 across different initial candidate numbers, and in ovarian and colorectal cancers it retains epithelial and multiple microenvironmental domains.

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