seurat `<obj>@assays$RNA@counts` vs `<obj>@assays$RNA@data`?

Question

[mk]

I have two matrices called <object>@assays$RNA@counts and <object>@assays$RNA@data that are both real non-negative.

What is the difference between these?

Answer 1

[rpolicastro]

@data will hold the normalized counts if you run NormalizeData. @counts will always hold the raw counts.

[mk]

Hi rpolicastro , I noticed that:

> min(<object>@assays$RNA@data)
[1] 0

Shouldn't this be negative assuming that seurat still normalizes to zero-centered values?

[rpolicastro]

The ScaleData function scales and centers the normalized data. I believe it goes to the @scale.data slot though and not @data.

[mk]

oh, that's interesting, so would <object>@assays$RNA@data just be linear in some previously-computed cell-wise size-factors (i didn't know seurat did that so generally use scran to compute size factors separately)?

if so then that matrix is very useful for some other stuff, too

[rpolicastro]

NormalizeData divides each feature in a cell with the total UMI counts for that cell, multiples the number by 10000, and then natural log + 1 transforms the counts, so you don't expect any negative values. ScaleData then centers and scales those counts.