Smartseq data prep for SCVI

Help scvi-tools
,
1

Hi all,

Does anyone have any advice on how best to prepare raw smartseq cellxgene matrix for SCVI?

Although the tutorials use a lot of smartseq datasets, the preprocessing of such datasets is not mentioned.

Main questions are:

  • Should I normalise the data prior to use? e.g. norm for gene length, TPM etc.

Any help would be much appreciated.

2

Here we normalize the counts to the median gene length and round to integers.

1 Like
3

Thanks Adam,
I had missed this. Having tried this now, I was wondering if I could pick your brains about using scArches across assays.

Having reference mapped 2 smart-seq datasets to a core atlas of 10X data, I am noticing that smarteq data does not appear to map well (none of the cells are integrated).

Have you ever noticed this? I am wondering whether this is to do with the lack of a smart-seq dataset in the core atlas.

Any thoughts are much appreciated

4

Hmmm, I would assume that global library size after the scaling might still be an issue. I would double check if they did anything else in the scIB paper as well.

To be honest, I haven’t done much reference mapping across technologies myself.