bc.pl.update_qualitative_palette is not working proprely

[ajulienla]

The combination dict is not taken in properly. Colors/labels are not matching as they should

ECM:

test = bc.datasets.Kotliarov2020_processed()
sc.pl.umap( test, color = ['celltype2'] )
import seaborn as sns
uniqueValues = (test.obs.get( "celltype2").unique().tolist())
yy = sns.color_palette("tab10", len(uniqueValues), as_cmap= False).as_hex()

## SHOULD BE ONE AND THEN THE REVERSE SCALE

color_palette = {i:j for i, j in zip(uniqueValues, yy)}
color_palette2 = {i:j for i, j in zip(uniqueValues[::-1], yy)}

bc.pl.update_qualitative_palette(test, color_palette, "celltype2" )
sc.pl.umap( test, color = ['celltype2'] )
bc.pl.update_qualitative_palette(test, color_palette2, "celltype2" )
sc.pl.umap( test, color = ['celltype2'] )

## CHECKING COLORS VSALUES SHOWS THAT THE KEY -VALUE RELATHION SHIP IS NOT PRESERVED
print(color_palette)
print(color_palette2)

[ajulienla]

Waiting for scanpy update as it seems that there has been some changes there too.

[kohleman]

The root cause of this is that the mapping between categories and colors is not read from a single dictionary in scanpy. The categories (e.g. 'B cell') are read from the AnnData object (pd.Categorical(adata.obs["celltype2"]).categories) and the colors can be defined in a separate list (adata.uns[group + "_colors"]). Although you can change the colors in the AnnData object, the order of the of the categories stay the same. For details have a look here:
scanpy/plotting/_tools/scatterplots.py: _get_palette (crruently line 1175): Here is the mapping done between category and color: dict(zip(values.categories, adata.uns[color_key])) with values = pd.Categorical(adata.obs[values_key]). I will re-write the besca/pl/_update_palette.py function to be able to accept an input. For testing this I replaced the following line:

uniqueValues = test.obs.get("celltype2").unique().tolist()

with

uniqueValues = pd.Categorical(test.obs["celltype2"]).categories.tolist()

This ensures that we use the same mechanism as scanpy uses to determine the categories and preserving the order