basic support for AnnData accessors

[ilia-kats]

The main point of this is to still be able to pass MuData objects to scanpy's plotting functions. This is currently possible due to MuData's implementation of obs_vector and var_vector, but if I understand the scanpy changes correctly (e.g. here) this will stop working as soon as anndata 0.13 is released.

The logic is based on the current _attr_vector implementation, which will go away at some point when obs_vector and var_vector are removed.

I'd appreciate some feedback on where else scanpy has switched to accessors. For example, it is currently possible to run sc.pp.neighbors and sc.tl.umap on a MuData object. Have these also switched to using accessors to access .obsm/obsp?

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 91.48%. Comparing base (1837822) to head (ac1fbd2).

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #147      +/-   ##
==========================================
+ Coverage   91.38%   91.48%   +0.10%     
==========================================
  Files          10       10              
  Lines        1601     1620      +19     
==========================================
+ Hits         1463     1482      +19     
  Misses        138      138              

Files with missing lines	Coverage Δ
src/mudata/_core/mudata.py	93.06% <100.00%> (+0.15%)	⬆️

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[ilan-gold]

Let's wait for @flying-sheep to get back to review this. There may be a cleaner way to this aside from violating types

[ilan-gold]

So this relies on the fact that MuData carries an obs, var etc. but does not follow the type?

https://anndata.readthedocs.io/en/latest/generated/anndata.acc.RefAcc.html#anndata.acc.RefAcc.get

[ilan-gold]

Maybe we should use a protocol?

[ilia-kats]

So this relies on the fact that MuData carries an obs, var etc. but does not follow the type?

Exactly.

[ilia-kats]

There may be a cleaner way to this aside from violating types.

You're violating types every time you pass a MuData to scanpy, everything there is typed as AnnData. Perhaps a protocol would indeed make sense.

[ilan-gold]

You're violating types every time you pass a MuData to scanpy, everything there is typed as AnnData. Perhaps a protocol would indeed make sense.

But maybe there's a way we can stop doing this now that scanpy is being redesigned - perhaps with a view to MuData.