Color scale not consistent between plots

[bengouts]

Hi all,

I observed a very misleading behaviour with the color scale : the default color scheme is different from one plot to another, which is extremely misleading.

h5path = '/home/user/Download/4-cell-lines-AML-multiomics.dna+protein.h5'
sample = ms.load(
    h5path,
    raw=False,
    filter_variants=True,
    filter_cells=False,
    single=True
)
sample.protein.normalize_reads(method='NSP')
sample.protein.run_umap(attribute='normalized_counts', output_label="UMAP_PROT")
sample.protein.cluster(attribute='normalized_counts', method='graph-community', k=100)

Then I plot the heatmap :

sample.protein.heatmap(attribute='normalized_counts')

Then I plot the UMAP :

sample.protein.scatterplot(attribute='UMAP_PROT', colorby=sample.protein.get_labels())

For instance, Cluster 1 is blue in the heatmap and red in the UMAP, which is extremely misleading.

Thanks for this very usefull and well documented library. Best regards, Benoit

[KKJSP]

@bengouts To ensure that the same colors are used in the scatterplot you have to pass "label" to colorby instead of the labels.

sample.protein.scatterplot(attribute='UMAP_PROT', colorby="label")

Quoting the documentation:

In case ‘label’ is provided then the stored paltte is used. If the values are strings, then a discrete color map is assumed. For numerical values a continuous color scale is used.

[bengouts]

Oups, my mistake. I assumed it used the labels by default. Thanks!

[bengouts]

Just one other example to higlight that default behaviour of color scale is misleading :

sample.protein.cluster(attribute='normalized_counts', method='graph-community', k=100)
sample.protein.add_row_attr("cluster_graph_community", sample.protein.get_labels())
sample.protein.heatmap(attribute='normalized_counts', splitby=sample.protein.row_attrs["cluster_graph_community"])
sample.protein.scatterplot(attribute='UMAP_PROT', colorby=sample.protein.row_attrs["cluster_graph_community"])

[KKJSP]

As of now the only to ensure that all plots have the same color is to store the value in the label row attribute of the assay by using sample.protein.set_labels(sample.protein.row_attrs["cluster_graph_community"]) and then passing "label" to the colorby or splitby parameters. The colors can be modified using sample.protein.set_palette() as well. Whenever an array is passed to these parameters, the function has no way of knowing what that array was supposed to indicate.

However I agree that mosaic should use a consistent palette for any given array across all the plots. We can keep this ticket open.