Should exploratory spatial transcriptomics ever receive ROBUST maturity without orthogonal negative controls?

[HERRY423]

Background & Scientific Problem

Spatial transcriptomics methods (e.g., Squidpy Moran's I, SpatialDE, Seurat FindSpatiallyVariableFeatures) identify spatially variable genes (SVGs) and cellular neighborhood enrichments.

However, spatial data is plagued by:

• Tissue edge artifacts (cells near tissue boundaries having fewer neighbors).
• Spot cell density variations (spots with 15 cells vs 2 cells in 10x Visium).
• Optical vignette distortion in image-based platforms (Xenium, MERSCOPE).

BioNexus Proposed Policy (Flagship Capability C)

• Default spatial analysis (without null models): Capped at FRAGILE.
• Spatial analysis with coordinate permutation null model (permuted_coords_null): Advances to SUPPORTED.
• Spatial analysis with cross-slice replication & orthogonal negative controls: Advances to ROBUST.

Questions for Discussion

1. Do you agree that uncorrected spatial autocorrelation metrics should never exceed FRAGILE?
2. What is your preferred negative control for spatial transcriptomics (coordinate permutation, spot density regression, or synthetic random field nulls)?