FindMarkers on scATAC-seq using normalized Tn5 cuts?

[ccshao]

I have a quite naive but anyway would like to confirm that the FindMarkers in Signac using the TF-IDF normlized Tn5 Cuts counts from Peak/Cell matrix provided by 10x, but NOT based on the fragment file?

Actually I saw a related note here:
#373 (comment)

The reason why I asked is that I think ArchR is using the fragments as the basic mearements (converted to tilematrix) and then performed the TF-IDF normalization. Differentially accessible peaks are computated subsequently. In the vignette I didn't find how they use Tn5 cuts at all. To me Signac and ArchR are quite different from the very beginning.

These two approaches also give us quite different results on clustering an UMAP.

Thanks!

[timoast]

I think the issue you've linked to gives the answer to this question

Whether cut sites or fragments are counted to create the matrix depends on the method used. Cellranger counts cut sites, Signac currently counts fragments.

[ccshao]

Cellranger counts cut sites, Signac currently counts fragments.

Now I get more confused. Wouldn't the data used in Signac be the cut sites from Peak/Cell matrix?
I mean, GetAssayData(object = pbmc, slot = "counts") will give the cuts in peak region?

[timoast]

I'm referring to the FeatureMatrix() function in Signac counting fragments rather than cut sites