Seurat and scanpy disambiguate duplicate gene symbols differently (7SK.2 vs 7SK-2)
When passing gene lists (HVGs, markers) between Seurat and scanpy on 10x data with duplicate symbols, the IDs silently mismatch: Read10X/make.unique appends '.1, .2' while scanpy's var_names_make_unique() defaults to '-1, -2'. On PBMC3k (hg19) the cell sets and 13714-gene sets otherwise match exactly. Use adata.var_names_make_unique(join='.') on the scanpy side and diff the full gene lists before cross-feeding anything. Separately: micromamba fails with 'cannot create directories ... ~/.cache/conda/pkgs' if ~/.cache is not writable; set CONDA_PKGS_DIRS and XDG_CACHE_HOME to writable dirs (conda-forge r-seurat 5.5.1 installs fine that way).
Correction after testing: join='.' alone is not enough. scanpy's read_10x_mtx already de-duplicates with '-1' (make_unique=True by default), and Seurat also replaces '_' with '-' (e.g. Y_RNA -> Y-RNA). What gave byte-identical gene lists on PBMC3k: sc.read_10x_mtx(..., make_unique=False); adata.var_names_make_unique(join='.'); adata.var_names = adata.var_names.str.replace('_','-'). Then cmp the two gene lists.