GLN_HIP chemical perception fails

[j-wags]

(off-dev) jeffreywagner@JW-MBP$ python test_parameterize_tripeptides.py

MainChain/GLN_HIP/GLN_HIP HarmonicBondForce 6.00869083404541 kJ/mol HarmonicAngleForce 101.89446258544922 kJ/mol PeriodicTorsionForce 134.29551696777344 kJ/mol NonbondedForce -252.6314697265625 kJ/mol -10.43280029296875 kJ/mol [H]C1=C(N(C(=[N+]1[H])[H])[H])C([H])([H])C@@(C(=O)N([H])C([H])([H])[H])N([H])C(=O)C@(C([H])([H])C([H])([H])C(=O)N([H])[H])N([H])C(=O)C([H])([H])[H] Traceback (most recent call last): File "test_parameterize_tripeptides.py", line 100, in off_sys = ff.create_openmm_system(mol.to_topology(),)#allow_nonintegral_charges=True) File "/Users/jeffreywagner/projects/OpenForceField/openforcefield/openforcefield/typing/engines/smirnoff/forcefield.py", line 1293, in create_openmm_system parameter_handler.create_force(system, topology, **kwargs) File "/Users/jeffreywagner/projects/OpenForceField/openforcefield/openforcefield/typing/engines/smirnoff/parameters.py", line 2520, in create_force exception_cls=UnassignedAngleParameterException, File "/Users/jeffreywagner/projects/OpenForceField/openforcefield/openforcefield/typing/engines/smirnoff/parameters.py", line 2197, in _check_all_valence_terms_assigned raise exception openforcefield.typing.engines.smirnoff.parameters.UnassignedAngleParameterException: AngleHandler was not able to find parameters for the following valence terms:

• Topology indices (33, 35, 36): names and elements ( C), ( N), ( H),
• Topology indices (36, 35, 37): names and elements ( H), ( N), ( C),

[j-wags]

It's the CNH angles about this atom. My first guess is that it's an aromaticity perception problem. I'll look into that possibility first.

[j-wags]

Second guess is that it's something to do with HIP being a superstructure of HID/HIE, and that this can be solved with a parameter-precedence change.

[j-wags]

Hm, looking at ALA_HIP shows a different pattern of double bonds on the sidechain. So it's possible that the charge moved. Yuck.

I need to figure out which structure was used for the training, and whether pymol is guessing at the double bonds or if they're explicit.

[j-wags]

The sidechain double bonds seem to be stated explicitly in the mol2 file. So it is an issue with the file.

I've just checked all XXX_HIP files, and only GLN_HIP has this bonding pattern.

[j-wags]

I suspect that this is a difference between the particular kekule structure in the dipeptide structure (which was used to make the FF/SMARTS) and the many kekule structures in the tripeptides involving HIS/HIE/HID/HIP. I could probably use a similar mechanism as #29 to fix this.