A way to visualize Molecular graph

[zarkoivkovicc]

Is there a way to visualize molecular graph? I am trying to debug a case where get_bond_rearrangs fails to determine the bond rearrangment due to wrong infererenced connectivity from xyz files. When I visualize the same xyz files in Avogadro2, the Avogadro infers correct bonds. The networkx draw function is completly useless for this kind of visualization, because it produces a messy drawing of a graph where a lot of nodes almost overlap. Rdkit also determines bond connectivity from xyz files correctly, so it must be a problem in the inference of bonds in autode.

[t-young31]

I'm not sure there is a good way unfortunately. Probably fastest to set the graph with a larger tolerance and look at the difference
i.e.

>>> import autode as ade
>>> m = ade.Molecule(atoms=[ade.Atom("H"), ade.Atom("H", x=1)])
>>> m.graph
MolecularGraph(|E| = 0, |V| = 2)
>>> from autode.mol_graphs import make_graph
>>> make_graph(m, rel_tolerance=0.4)
>>> m.graph
MolecularGraph(|E| = 1, |V| = 2)

if you're using metals you might need allow_invalid_valancies=True. Signature

autodE/autode/mol_graphs.py

Lines 128 to 133 in 6a6eebe

	def make_graph(
	species: "Species",
	rel_tolerance: float = 0.3,
	bond_list: Optional[List[tuple]] = None,
	allow_invalid_valancies: bool = False,
	) -> None:

[zarkoivkovicc]

Rdkit has a very nice function to determine molecular graph (including bond orders) from xyz coordinates. Is there any specific reason why autode doesn't use that instead of primitive algorithm based on covalent radii?

[t-young31]

when I tried it it wasn't any good, but that was probably 4 years ago now so many times have changed! what does RDKit do if not something based on covalent radii?

[shoubhikraj]

@zarkoivkovicc I have used the XYZ reader in RDkit before. From Greg Landrum's presentation here it seems that RDKit's XYZ reader uses xyz2mol. Which also uses a covalent radii based cutoff. The only difference is the algorithms used for determining bond orders, which seems to be more sophisticated than the one we use for autodE. I don't know if they updated their algorithm in later versions or not.

For determining bond rearrangements, the bond orders are not used in any case - so I don't think it will change things much. If it is metals, then its probably the valency pruning which is causing an issue as @t-young31 said - I have run into this before.

[zarkoivkovicc]

I am sorry, I only took a quick glance at xyz2mol code and I assumed based on the legth of the code it implements something more advanced, but it turned out all fancy logic is related to the bond order determination, not to connectivity inference. Moreover autodE uses the same relative tolerance as rdkit by default, so I examined the examples where I thought something went wrong with autodE and it turned out the optimization of the structure didn't end up where I expected it to go.

Many thanks for fast reply and help!

[shoubhikraj]

Looking at xyz2mol, they also have a huckel theory bond order based connectivity but it is not used by default. I have also been testing a similar bond order based method, but it is not complete enough to be implemented in autodE yet.