I think there is a limitation to the hdf5 parser where it only looks for "shallow" groups i.e., groups with only one level of dataset.

Some example data:

wget https://datasets.cellxgene.cziscience.com/4cc9a6b9-ae3a-4084-b3f0-f19578eb30ac.h5ad -O sc_dataset.h5ad

And then:

import virtualizarr

store = virtualizarr.parsers.HDFParser()("./sc_dataset.h5ad", object_store=obstore.store.LocalStore('./'))
zarr.open_group(store, mode="r", zarr_format=3) # evidently I need to specify v3 otherwise this crashes without finding an group?

outputs

<Group ManifestStore(group=
ManifestGroup(
    arrays={},
    groups={},
    metadata=GroupMetadata(attributes={'encoding-type': 'anndata', 'encoding-version': '0.1.0'}, zarr_format=3, consolidated_metadata=None, node_type='group'),
)
, stores=<virtualizarr.manifests.store.ObjectStoreRegistry object at 0x15b4e2ab0>)>

which lacks the nested group structure:

import h5py

with h5py.File("./sc_dataset.h5ad") as f:
    print(list(f.keys()))

['X', 'layers', 'obs', 'obsm', 'obsp', 'raw', 'uns', 'var', 'varm', 'varp']

I think the culprit is here

This function is called in HDFParser but only handles level-0 datasets and not nested groups.

Maybe this is intentional, not sure! Thanks for the library, very cool!