Applying ssGSEA on Spatial Transcriptomics Visium data

[han9978]

Hi, thanks for the great tool. I was wodering, if there is there anything against applying the ssGSEA method to Spatial Transcriptomics Visium data? I proceeded as follows: I normalized the expression matrix, extracted it from Seurat object and applied the method:

mat<-NormalizeData(mat, normalization.method = "LogNormalize", scale.factor = 10000, assay="Spatial")
mat<-as.matrix(GetAssayData(mat, assay="Spatial", layer="data"))

obj<-gsva(ssgseaParam(mat, gene_list), verbose=T)

I used version 1.52.3 of the GSVA package.
The results seem to make sense but I would really appreciate your feedback on the use of ssGSEA on visium data or where problems might lie. I found many other papers where ssGSEA was applied on spatially variable genes or marker genes but none using it on the whole expression matrix.
Thank you

[rcastelo]

Hi, we are currently investigating this on the GSVA method, which is similar to the ssGSEA method. Currently, the GSVA method, i.e., more concretely the gsvaParam() parameter constructor function, can take as input a SpatialExperiment, but as you are doing in your code, it still does not exploit the spatial information during score calculations, i.e., taking the spots as independent columns/samples/cells. We are evaluating whether this is necessary or it requires other treatments up or downstream. In our hands, GSVA scores seem to give also sensible results, so the question seems to be more on whether exploiting the spatial information could improve them. Could you share some of the papers you found where ssGSEA was applied on spatial data?