Allow using arbitrary vectors in plotting (not only `obs` columns)

[picciama]

• Additional function parameters / changed functionality / changed defaults?
• New analysis tool: A simple analysis tool you have been using and are missing in sc.tools?
• New plotting function: A kind of plot you would like to seein sc.pl?
• External tools: Do you know an existing package that should go into sc.external.*?
• Other?

Quite often I need to color UMAPs based on features that are not part of adata.X but adata.obsm for the reason that they are special. E.g. KO data with gRNAs versus endogenes/ target genes, or viral genes versus edogenes.

Example use case:

• Cluster cells based on endogenes
• UMAP and color by a bunch of viral genes

Clustering must not include these viral genes -> must be excluded from X.
I don't want to store so many additional columns in obs and I need to have these features separated in their own matrix for downstream analysis, which is why I want to use obsm.

Can we have sth. like this:

sc.pl.umap(adata, color='viral_genes')  # adata.obsm['viral_genes'] is a pandas.DataFrame ?

It shouldn't be overcomplicated I think, since this only involves an additional check: if the elements in the color arg list are not found in obs.columns nor var.columns, then check the keys in obsm and use the entire dataframe behind this key.

[ivirshup]

This has been worked on here: scverse/anndata#342

The idea is to allow any vector from the anndata object to be used for coloring, but that PR seems a bit stalled at the moment. This would also be useful for providing parameters in other places, like regress_out.