Obs reassignment silently allows you to change obs_names

[Hrovatin]

When adding a whole obs to adata that already has obs_names the index matching is not checked. I would expect adata to complain if newly assigned obs has different index than obs_names.

>>> values = [0,1]
>>> values_rev = values[::-1]
>>> a = sc.AnnData(pd.DataFrame(np.array(values).reshape(-1,1), index=[str(i) for  i in values]))
>>> print('X original\n', a.X)
X original
 [[0.]
 [1.]]
>>> print('obs names original',a.obs_names)
obs names original Index(['0', '1'], dtype='object')
>>> o = pd.DataFrame(np.array(values_rev).reshape(-1,1), index=[str(i) for  i in values_rev])
>>> a.obs = o
>>> # Obs names/order change after adding new obs, but X stays the same
>>> print('obs new\n',a.obs)
obs new
    0
1  1
0  0
>>> print('X new\n',a.X)
X new
 [[0.]
 [1.]]

[ivirshup]

Thanks for suggestion. Could you give an example of a use case where this causes problems?

---

I would broadly agree that this is weird. It makes more sense for it to be the obs_names, var_names of the AnnData that the obs.index, var.index.

However, we can't control or see when a user does:

adata.obs.index = ...
adata.obs.set_index(..., inplace=True)

I would suspect the two examples I showed are common, and I'm not sure if we can disallow them easily.

[Hrovatin]

So I made an adata without obs (just X with names in var and obs, from an expression data frame). Then at latter point I wanted to add a whole obs df, e.g. adata.obs=df, but this would cause index that is already in adata from creation to change. I would expect that if new var or obs would be assigned to adata the adata would check if index in adata matches index in the new df.

[ivirshup]

What was the index in the new dataframe? Was it the right labels, but in the wrong order? Or a totally different set of labels?

[Hrovatin]

It was right labels in a wrong order. But I would expect an error if totally wrong index and for wrong order not sure if error or reordering (as in pandas)

[jeanettejohnson]

an example of when this would be an issue is when you're plotting cell data colored by annotation. if the annotation doesn't correspond to the cell it becomes meaningless.

[github-actions[bot]]

This issue has been automatically marked as stale because it has not had recent activity. Please add a comment if you want to keep the issue open. Thank you for your contributions!