Can't save MuData object to h5mu file due to OrderedDict

[PauBadiaM]

Hi,

When trying to write the mudata object, h5py crashes due to some .uns values being OrderedDict instead of normal dictionaries.

TypeError: No method has been defined for writing <class 'collections.OrderedDict'> elements to <class 'h5py._hl.group.Group'>

Here is how to reproduce the bug:

import muon as mu
import os


# Read and create mdata object
data_dir = 'pbmc10k/'
mdata = mu.read_10x_h5(os.path.join(data_dir, "filtered_feature_bc_matrix.h5"))
mdata.var_names_make_unique()

# Write
mdata.write("test.h5mu")

The pbmc10k is the data used in this tutorial.

Apparently, when reading the inputs, mu.read_10x_h5 creates two OrderedDicts in mdata.mod['atac'].uns. If I set them to dict then it saves the object correctly but it would be nice to handle this automatically.

mdata.mod['atac'].uns['files'] = dict(mdata.mod['atac'].uns['files'])
mdata.mod['atac'].uns['atac'] = dict(mdata.mod['atac'].uns['atac'])

Thanks you for your time!

System

• OS: Ubuntu 22
• Python 3.8.13
• anndata 0.8.0
• mudata 0.2.0
• muon 0.1.3

[gtca]

Hey @PauBadiaM,

Thanks for opening this issue, we've been moving away from OrderedDict in muon but I realise that the combination of older versions of muon and newer versions of anndata results in this error...

from collections import OrderedDict
import numpy as np
from jax import random
from anndata import AnnData

x = random.normal(key=random.PRNGKey(1), shape=(1000,500))
adata = AnnData(np.array(x))

adata.uns["ord_dict"] = OrderedDict({"key1": "value1"})
adata.write("anndata_with_ordered_dict.h5ad")
# => TypeError: No method has been defined for writing <class 'collections.OrderedDict'> elements to <class 'h5py._hl.group.Group'>

We aim to circumvent it in two ways: use dict() in muon from now on, as mentioned above, as well as see if we can account for OrderedDict in AnnData's I/O spec. For the latter, I've just opened scverse/anndata#796.

[bioarpit1]

Any solution to this ?

[gtca]

@bioarpit1 This has to be implemented on the AnnData's side: scverse/anndata#796.

Here, the solution is simply converting OrderedDict to dict where this issue arises.