MOL2Parser requires ATOM optional fields

[RMeli]

Expected behavior

I expect MDAnalysis to be able to parse standard MOL2 files.

Actual behavior

If I understand correctly the MOL2 standard, a <TRIPOS>ATOM data line only requires the 6 following fields: atom_id atom_name x y z atom_type. However, the MOL2Parser expects also the following optional fields to be present [subst_id [subst_name [charge [status_bit]]]] (actually, only the first three; see #2318).

Code to reproduce the behavior

from io import StringIO

import MDAnalysis as mda

print(mda.__version__)

mol2="""\
@<TRIPOS>MOLECULE
MOL2
2
SMALL
NO_CHARGES


@<TRIPOS>ATOM
  1 N1       6.8420     9.9900    22.7430 N.am
  2 N2       4.4000     9.1300    20.4710 N.am
"""

u = mda.Universe(StringIO(mol2), format='MOL2')

2.0.0-dev0
Traceback (most recent call last):
  File "/Users/rmeli/Documents/git/software/mdanalysis/package/MDAnalysis/core/universe.py", line 122, in _topology_from_file_like
    topology = p.parse(**kwargs)
  File "/Users/rmeli/Documents/git/software/mdanalysis/package/MDAnalysis/topology/MOL2Parser.py", line 153, in parse
    aid, name, x, y, z, atom_type, resid, resname, charge = a.split()[:9]
ValueError: not enough values to unpack (expected 9, got 6)

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "test2.py", line 20, in <module>
    u = mda.Universe(StringIO(mol2), format='MOL2')
  File "/Users/rmeli/Documents/git/software/mdanalysis/package/MDAnalysis/core/universe.py", line 338, in __init__
    **kwargs)
  File "/Users/rmeli/Documents/git/software/mdanalysis/package/MDAnalysis/core/universe.py", line 140, in _topology_from_file_like
    "Error: {2}".format(topology_file, parser, err))
ValueError: Failed to construct topology from file None with parser <class 'MDAnalysis.topology.MOL2Parser.MOL2Parser'>.
Error: not enough values to unpack (expected 9, got 6)

Current version of MDAnalysis

• Which version are you using? 2.0.0-dev0
• Which version of Python? ython 3.7.10
• Which operating system? macOS 11.5.1

Additional Information

This is linked to #2318 and my fix in #2319 was only partial. Other fields should be optional (especially important if NO_CHARGES in @<TRIPOS>MOLECULE is present).

The problematic line is the following:

mdanalysis/package/MDAnalysis/topology/MOL2Parser.py

Line 140 in 09f50ec

aid, name, x, y, z, atom_type, resid, resname, charge = a.split()[:9]

[orbeckst]

If no charges are present then I would expect the topology to not contain charge information. This is a much safer approach than trying to guess charges, which we cannot do well anyway. If the topology does not contain charges it will simply not have methods such as total_charge() but everything else (EDIT: that does not require charge information) will work.

@RMeli @IAlibay @richardjgowers @lilyminium does this sound sensible?

[IAlibay]

Not read the full context of this discussion, but yeah charge is an optional topology attribute, if there's no charge then we shouldn't guess it (which would require rdkit anyways)

[orbeckst]

Following up on #3598 (comment) : I think resname and resid are always required in our topology system so using a default of "UNK" and 1 seems reasonable to me. (But I am happy to be corrected!)

[Mrqeoqqt]

Hi @IAlibay Should I discard the charge attribute when any of the charge is empty? Is there any possibility that the part of atoms in .mol2 file have charges but the others don't?

[RMeli]

If no charges are present then I would expect the topology to not contain charge information.

Same, this is what I was trying to say in #3598 (comment)

I think resname and resid are always required in our topology system so using a default of "UNK" and 1 seems reasonable to me.

I was not aware of this. If they are required fields, then the chosen default values are very sensible. But charges should be discarded (especially since they are optional).

[orbeckst]

On discord, @richardjgowers said:

currently required are resnums, resids and segids, but not resnames. XYZParser is good to look at since it's the most minimal data parser: https://github.com/MDAnalysis/mdanalysis/blob/develop/package/MDAnalysis/topology/XYZParser.py
Hmm but strangely Topology never enforces this minimum set, but I'm imagining the selection language and elsewhere assumes they exist... A bit of a grey area I'll admit

@lilyminium pointed to the list of canonical attributes and confirmed that resname is not essential.

This means that

• resid = 1 is a good default
• resname can be left empty and then will not be included in the Topology; do not set it to a default value
• charges are optional so if not present, do not include in Topology (just as @RMeli said)

[orbeckst]

More discussion (#3598 (comment)): At the moment I am proposing to fail with a ValueError if some lines contain optional data (such as a resname, aka subst_name) and others don't. From the definition for @<TRIPOS>ATOM

atom_id atom_name x y z atom_type [subst_id [subst_name [charge [status_bit]]]]

it looks as if this would be allowed but then things become difficult for our topology object and we would have to guess or set a resname for some atoms but not others, would need to check that a default resid (subst_id) is compatible with what we guessed, or for charge, would need to set some charges to 0 or guess while others were read from the file. I'd much rather have consistent handling of all optional data: If optional data values are present for one atom then they need to be present for all of them.

[RMeli]

I'd much rather have consistent handling of all optional data: If optional data values are present for one atom then they need to be present for all of them.

I totally agree with this.