FindSpatialVariableFeatures slot options

[kris9393]

Hi,

When using the FindSpatiallyVariableFeatures function to identify genes which exhibit spatial clustering in Visium 10X samples, I was wondering which slot is best to use after using running SCTransform?

brain <- FindSpatiallyVariableFeatures(brain, assay = "SCT", features = VariableFeatures(brain)[1:1000],
selection.method = "moransi")

The default slot seems to be “scale.data” if the argument is left unspecified, but I was wondering whether the “data” or “counts” slot makes more sense to use for calculating Moran’s I and comparing values between samples. Do you have any advice on which slot to use and why?

Thanks.

[YoukaiFromAccounting]

Hi @kris9393; thank you for your question and apologies for the delay in response.

When using FindSpatiallyVariableFeatures() with data processed through SCTransform() (which is typically used on a Seurat object prior to finding spatially variable features), we recommend using the scale.data slot as this slot contains the Pearson residuals from the regularized negative binomial regression used by SCTransform. This model specifically corrects for technical variation (ie: sequencing depth) while preserving biological signal; as such, they are suitable for comparing values within and between samples when calculating spatial autocorrelation metrics like Moran's I.