Topology groups and PDB files

[mimischi]

Just stumbled upon the problem, that accessing any method that involves topology groups does not work, when creating a Universe with a PDB or GRO file. Both the master version 0.15.0, as well as a freshly cloned develop version 0.16.0-dev do not work, while producing different errors.

It does work when loading the topology from the test suite, so I'm using that as the expected behaviour. An non-working example PDB file would be 1UBQ from PDB.

Expected behaviour

from MDAnalysis.tests.datafiles import PSF
u = mda.Universe(PSF)
u.dihedrals
<TopologyGroup containing 8921 Dihedrals>

Actual behaviour

MDAnalysis 0.15.0

u = mda.Universe('1ubq.pdb')
u.dihedrals
<TopologyGroup containing 0 Nones>

MDAnalysis 0.16.0-dev

u = mda.Universe('1ubq.pdb')
u.dihedrals
---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-45-133525c932b4> in <module>()
      1 c = mda.Universe('1ubq.pdb')
----> 2 c.dihedrals

~/Code/mdanalysis/package/MDAnalysis/core/universe.pyc in dihedrals(self)
    516     def dihedrals(self):
    517         """Dihedral angles between atoms"""
--> 518         return self.atoms.dihedrals
    519 
    520     @property

~/Code/mdanalysis/package/MDAnalysis/core/groups.pyc in __getattr__(self, attr)
   1503                 pass
   1504         raise AttributeError("{cls} has no attribute {attr}".format(
-> 1505             cls=self.__class__.__name__, attr=attr))
   1506 
   1507     def __reduce__(self):

AttributeError: AtomGroup has no attribute dihedrals

The .angles method does not work either.

Looking through the issues I came across #1253, which sounds pretty familiar.

[richardjgowers]

Hey, thanks for the report

So a pdb/gro doesn't technically define dihedrals or angles, but obviously they can be inferred from bonds. So in a way this is "working as intended".

There's currently u.atoms.guess_bonds() which will try to guess bonds (based on distances and guessed vdw radii), then angles, then dihedrals. If you call this explicitly you should get the bonds from your PDB file. (Assuming the vdw radius can be guessed for all atoms).

If you load a PDB with CONECT records, then you can't immediately use them to guess angles admittedly, maybe there should be a u.atoms.guess_angles() method too..

[mimischi]

I see, you have to call u.atoms.guess_bonds() explicitly or set guess_bonds=True for the Universe object. Thanks!

Wouldn't it be beneficial to catch that exception in the current develop version?

[richardjgowers]

Yeah the error message could definitely be better, like providing a customised AttributeError for a known list of things

[mimischi]

I'd be happy to contribute if that's ok!

[richardjgowers]

@mimischi yeah that would be great! I think if you extended the __getattr__ here to add a special case for bonds etc. If you're doing that, you should also write a test that checks the errormessage

[mimischi]

Finally coming back to this issue, I noticed the depreciation warnings and #1377.

Does the issue now shift from AtomGroup.__getattr__ to Atom.__getattr__?

[mimischi]

Ping 👋

[kain88-de]

ping @richardjgowers can you comment on this?