Including "counts" dataset in conversion to AnnData

[roussine]

Hello all -

as I can see, the Convert function of SeuratDisk (from h5Seurat to h5ad) will never touch the counts dataset (not-normalised matrix) of the default assay (e.g., "RNA") in a h5Seurat object, as it gives priority to "scale.data" (variable features only) and then "data" (normalised matrix) when filling X.

Is there a way to force such conversion?.. Or maybe use an alternative to SeuratDisk (as no longer supported..)? sceasy seems not a solution, as it picks only a normalised matrix and ignores the rest.

Thanks a lot in advance for any help
Leo

[ccnawrocki]

I see this is from awhile ago, but I just did it by doing this in R:

obj@assays$RNA@data <- obj@assays$RNA@counts
SaveH5Seurat(object = obj, filename = "obj_rawcts.H5Seurat", overwrite = T) # Using SeuratDisk package
Convert(source = "obj_rawcts.H5Seurat", dest = "obj_rawcts.h5ad", overwrite = T) # Using SeuratDisk package

Then in Python:

obj = sc.read_h5ad("obj_rawcts.h5ad")

Now, the anndata object has the counts in obj.raw.X. This will be float, so you'll want to convert to integer before doing anything else. You can just reproduce the normalized data by re-doing the normalization step with scanpy or by doing it manually (pretty feasible, if you had done log norm in Seurat).

Maybe this will help someone else down the line.