With all the mouse data, ~80 batches, doublet detection produces too many figures. Should save instead of plot them.
anndata.Concatenate() uses >100GB of memory. Use in "inner" join instead of an "outer" join may solve this, but at the risk of discarding genes.
see the mouse branch for ongoing experiments: https://gitlab.unige.ch/nef_lab/public_scrnaseq/-/tree/mouse