Gregor Sturm

Should get be a module, or a function? I think `ir.get` as a module is flexible and future-proof. There might be more functions added in the future. Even if there...

For easy plotting with scanpy functions, we could do one of the following. 1. Nothing. The user is free to assign a variable of interest to a column ```python adata.obs["locus_vj1"]...

I like options (1) and (3), and they are not mutually exclusive. (2) seems like it's to inflexible and leads to too many special cases.

Anndata supports now categoricals with `nan` values. There are still some issues with anndata saving booleans. Also anndata can't save object dtypes, so we need to make sure to convert...

Anndata now also supports nullable booleans: https://github.com/theislab/anndata/issues/504#issuecomment-1010352216 Nullable strings are still not supported (https://github.com/theislab/anndata/issues/679), but less important since categoricals are better in almost all cases anyway.

superseded by #327

Hi @naity2, thanks for raising this issue. I felt the "populations of interest" would usually be clusters, and, therefore somewhat similar to "similarity in terms of gene expression". I'm interested...

Thanks for the response, especially the comparison by conditions that are independent of gene expression is a use-case I have overlooked. Do you have any thoughts on how to best...

Cool, thanks for your input! I'll remove the deprecation warning in the next update and think about how to streamline plotting.

MMSeqs2 is likely faster when ran on all sequences simultaneously and should produce the same results when ran with `mmseqs align`: https://github.com/soedinglab/mmseqs2/wiki When used with `mmseqs2 prefilter` it's even faster...