Regressing out genes

[joshuahwu]

In Seurat, there is an option of selecting a list of genes in the pre-processing regressOut function. I was wondering if there was similar functionality in Scanpy. Doing something like below does not work for me as a lot of the cells have 0 expression, giving me a PerfectSeparationError.

adata.obs[gene] = adata[:, adata.var_names==gene].X
sc.pp.regress_out(adata,gene)

Any help would be appreciated. Thank you!

[LuckyMD]

Hi,
I'm not entirely sure how Seurat does it, but I assume you could take the mean expression level (or mean z-score) of a couple of genes, store that in a .var column, and regress that out by sc.pp.regress_out(adata, var_col)?

Something like this:

adata.var['genes_of_interest'] = adata.X[:,gene_list].mean(0)
sc.pp.regress_out(adata, genes_of_interest)

If you want to ensure an equal contribution of all the genes to the gene score without weighting by mean gene expression, you could first use sc.pp.scale() on a copy of the adata object like this:

adata_tmp = adata.copy()
sc.pp.scale(adata_tmp)
adata.var['genes_of_interest'] = adata_tmp.X[:,gene_list].mean(0)
sc.pp.regress_out(adata, genes_of_interest)

Note that I have not tested this code... so no guarantees ;).

[ivirshup]

This would be very easy to implement with: sc.get.obs_df(adata, keys).

That would not solve the problem of (I believe) constant genes giving errors.