MOFA analysis with missing modalities and sparse matrices fails with ValueError: setting an array element with a sequence

[VPetukhov]

The bug description
I have multiple samples, some of them have CITE-seq data, while others only have scRNA-seq data. Integrating them with MOFA and use_obs='union' fails:

mu.tl.mofa(mdata_joint, groups_label=BATCH_VAR, use_obs='union')

Error:

162         ix = expanded.join(orig, how="right").ix.values
164         # set expanded data with part of the orig data
--> 165         x[ix, :] = mdata[m].X
166         data[i] = x
168 # Subset features if required

ValueError: setting an array element with a sequence.

The reason
It all comes from this line. The problem is that mdata[m].X is sparse, while x[ix, :] is dense. Probably, the same logic as before should be applied here.

System

• OS: macOS Ventura 13.3.1, M1
• Python version: 3.9.16
• Versions of libraries involved: muon==0.1.3

[gtca]

Thanks! Fixed as suggested, and I added a small test with this use case as well.