Replaces original da-barplot with ANCOMBC2-visualizer

[Macabe222]

Description

Adds transformer of type DataLoafPackageDirFmt to type ANCOMBC2OutputDirFmt allowing ancombc2_visualizer to accept outputs from ancombc. Renames ancombc2_visualizer to da_barplot and deletes original da_barplot.

Addresses #156

AI Disclosure

• NO AI USED.
• AI USED.

AI Usage Details

[lizgehret]

Hey @Macabe222, let us know when this one needs a review!

[Macabe222]

@lizgehret this one should be good now

[lizgehret]

Hey @Macabe222, this all generally lgtm; just some test coverage changes mentioned below!

[lizgehret]

It's probably worth adding some additional tests to show that both formats can be used for this visualizer. I also think the tests that were removed with the deletion of test_diff_abundance_plot should be moved/refactored into here so that we aren't just losing all of that test coverage for the visualizer that's now supporting double the inputs/outputs.

[lizgehret]

One other drive-by thought - maybe all of the shared tests for this visualizer should get moved into test_diff_abundance_plot and then the only tests in this file are for the ancombc2 method.

[lizgehret]

I'm looking back through our release announcements for 2026.1 and 2026.4 and it doesn't look like we added a 'breaking changes' announcement regarding the removal of the ancombc2 visualizer... maybe good to discuss in this week's eng meeting on whether we should merge this for 2026.7 or hold off until 2026.10 (even though this would be a nice one to get squared away).

EDIT: discussed in eng meeting (21 April 2026); this was in our breaking changes for 2025.10

[lizgehret]

Hey @Macabe222 just a heads up, I've implemented a new PR template that's now connected to the lint action in CI for all of our distro plugins. I've retroactively added the template to this PR (since you opened it prior). You'll just need to select the appropriate AI disclosure checkbox & fill in details on that if AI was used. lmk if you have any questions!

[Oddant1]

The failure here is a bit strange. Haven't fully dug through the framework to figure out why exactly this is working in this way, but

Here we call ancombc2 as a raw function

cls.abc2_output = ancombc2(
    table=cls.biom_table,
    metadata=cls.ancombc2_metadata,
    fixed_effects_formula='body-site + year',
    group='body-site',
    structural_zeros=True
)

ancombc2 returns a value from transform here

return transform(data=slices, to_type=ANCOMBC2OutputDirFmt)

This is the transformer it is invoking.

@plugin.register_transformer
def _2(slices: ANCOMBC2SliceMapping) -> ANCOMBC2OutputDirFmt:
    '''
    Transforms a dataframe of ANCOMBC2 model statistics into the ANCOMBC2
    output directory format.
    '''
    format = ANCOMBC2OutputDirFmt()
    for slice_name, slice_df in slices.items():
        format_slice = format.__getattribute__(slice_name)
        format_slice.write_data(slice_df, pd.DataFrame)

    extra_files = ['diff', 'passed_ss']
    if not any(file in slices.keys() for file in extra_files):
        header = {
            "doctype": {
                "name": "table.jsonl",
                "format": "application/x-json-lines",
                "version": "1.0"
            },
            "direction": "row",
            "style": "key:value",
            "fields": [
                {"name": "taxon", "type": "string", "missing": False},
                {"name": "(Intercept)", "type": "number", "missing": False}
            ],
            "index": [],
            "title": "",
            "description": "",
            "extra": {}
        }

        for file in extra_files:
            output_path = os.path.join(format.path, file + '.jsonl')
            with open(output_path, 'w') as f:
                f.write(json.dumps(header, separators=(',', ':')) + "\n")

    return format

We run the following two tests in the following order (some other tests also run and aren't relevant to the failure):

First, test_da_barplot_action_ancombc2 which loads our abc2_out as an Artifact then calls the da_barplot Visualizer.

def test_da_barplot_action_ancombc2(self):
    pm = PluginManager()
    da_barplot = pm.plugins['composition'].actions['da_barplot']

    ancombc2_artifact = Artifact.import_data(
        type='FeatureData[ANCOMBC2Output]', view=self.abc2_output
    )

    viz = da_barplot(ancombc2_artifact)
    viz = viz.visualization

    with tempfile.TemporaryDirectory() as tempdir:
        viz.export_data(tempdir)
        assert os.path.exists(os.path.join(tempdir, 'index.html'))

Second, test_da_barplot_non_overlapping_taxonomy, which uses abc2_out, the Format that was returned from ancombc2 via the transformer and not the Artifact.

def test_da_barplot_non_overlapping_taxonomy(self):
    '''
    Tests that an error is raised when attempting to visualize ancombc2
    data with a taxonomy that contains none of the features present in the
    ancombc2 data.
    '''
    taxonomy_df = pd.DataFrame({
        'Feature ID': ['feat1', 'feat2'],
        'Taxon': ['taxon1', 'taxon2'],
        'Confidence': [0.9, 0.95]
    })

    with tempfile.TemporaryDirectory() as tempdir:
        with self.assertRaisesRegex(
            ValueError, 'No features remained in your taxonomy'
        ):
            da_barplot(tempdir, self.abc2_output, taxonomy_df)

    with tempfile.TemporaryDirectory() as tempdir:
        with self.assertRaisesRegex(
            ValueError, 'No features remained in your taxonomy'
        ):
            da_barplot(tempdir, self.abc_output, taxonomy_df)

I'm not sure I fully understand what's happening yet, but due to the invocation of the transformer at the end of ancombc2, the Format we created previously is deleted from the disk when we load it as an Artifact. I'm not positive why this doesn't happen when I return a Format from ancombc2 without invoking the transformer. Something to do with how the views are coerced.

@lizgehret @ebolyen thoughts? Do we want to ditch invoking the transformer at the end of ancombc2? I can look deeper through the framework and try to refactor it such that invoking the transformer doesn't cause this issue. I've looked at it a bit and there are of course parts in the transformer where we create different views and different objects. We are only loading the original path as read-only but something about the lifecycle of the view created is causing it to go away when loaded as an Artifact. That part I haven't really dug into yet.