Question regarding enrichIt

[ncborcherding]

Question regarding enrichIt

[ncborcherding]

Hey rythi4,

Would you mind sharing an example?

If you are using a single-cell object to run enrichIt(), the function is pulling the raw counts from the "RNA" assay and calculates the enrichment score using a Poisson distribution in GSVA, and performs the ssGSEA normalization described by Barbie et al 2009. This normalization uses the absolute difference between a given gene set - so it is occurring on a gene.set-level.

So if the composition of the seurat object changed, in terms of the cell number, there might be a difference in the output values. But if it is the same Seurat object, it should result in the same values.

Nick

[rythi4]

Here is an example:

`
#Making GeneSetCollections:
set1 <- GeneSetCollection(GeneSet(genes[["X"]][["gene"]], setName = "x"), GeneSet(genes[["Y"]][["gene"]], setName = "y"), GeneSet(genes[["Z"]][["gene"]], setName = "z"), GeneSet(genes[["W"]][["gene"]], setName = "w"))
set2 <- GeneSetCollection(GeneSet(genes[["X"]][["gene"]], setName = "x"), GeneSet(genes[["Y"]][["gene"]], setName = "y"))

#Here I define the seurat_object: seurat_object <- XXX
DefaultAssay(seurat_object) <- "RNA"
seurat_object_1 <- UpdateSeuratObject(seurat_object)
ES.seurat_1 <- enrichIt(obj = seurat_object_1, gene.sets = set1, groups = 1000)
#Here I define the seurat_object: seurat_object <- XXX
DefaultAssay(seurat_object) <- "RNA"
seurat_object_1 <- UpdateSeuratObject(seurat_object)
ES.seurat_2 <- enrichIt(obj = seurat_object_1, gene.sets = set2, groups = 1000)
`
The two seurat_objects are defined by the exact same line of code.

When I look at ES.seurat_1 and ES.seurat_2 (I order the columns 'x' by highest value for easy comparison), I see that the highest value is not the same:

[ncborcherding]

Thank you so much for the reproducible example!!

I found the issue - it lies in the GSVA package itself

I think this might be an error in my reading of the original manuscript for ssGSEA. I will take a look, I might have to implement an intrinsic normalization so the enrichment scores are deterministic. I'll post here when I get something working.

Thanks for bringing this to my attention.

Nick

[rythi4]

Of course, that was no problem!

[ncborcherding]

The newest dev version (v1.3.5) now has a new parameter ssGSEA.norm which will scale the columns from 0 to 1. However, if you would like to subset the data at a later time, the default setting is now not to normalize the data.

enrichIt <- function(obj, gene.sets = NULL, 
                     method = "ssGSEA", 
                     groups = 1000, 
                     cores = 2,
                     min.size = 5,
                     ssGSEA.norm = FALSE) 

Thanks again for letting me know about this.

Nick