What is the correct input count matrix (which assay/slot) to generate the adata object?

[reneemoerkens]

Hi scVelo development team,

Thanks for the great package!

I am using PAGA, scVelo and CellRank to analyse my scRNAseq dataset and I am a bit confused as to which count matrix I should use as input.
I have my preprocessed adata object, that was previously SCTransformed (v2) as Seurat object following the Seurat vignette. I will integrate my adata object with my unprocessed loom files (ldata object) to start trajectory analysis. To generate my adata object I have used the count matrix of the SCT assay in the data slot (log-transformed counts). After running the function 'scv.pp.filter_and_normalize', I get the message:
'WARNING: Did not normalize X as it looks processed already. To enforce normalization, set enforce=True.
Normalized count data: spliced, unspliced.'

Which count matrix data should I use as input for the adata object? The SCT assay data (log-transformed) or count (corrected) slot? Or the RNA assay data (normalised) or count (unnormalised) slot? Or does this not matter as mainly the unspliced/spliced variants are used from the loom files which are unnormalised when starting the trajectory analysis?

Thank you so much!