Furan inchi and SMILES cannot be resolved

[connie]

Due to lack of aromaticity perception molecules like Furan cannot be resolved using RMG-Py.
SMILES: c1ccoc1
InChI=1S/C4H4O/c1-2-4-5-3-1/h1-4H

[rwest]

What happens if you enter it via its adjacency list? Then I think we bypass OpenBabel.

[connie]

Yeah it works fine when you do that. When are we fully switching to RDKit? =p

[rwest]

As soon as you wish!
It can read and write InChI and SMILES, and canonical SMILES.
We'd lose CML from our input files, but I don't think that's a big cost.
I look forward to merging your rdkit branch, @connie 😜

[rwest]

Oh, one more thought - @pierrelb, can rdkit do the gaussian log file (i.e. coordinates only) to molecule conversion, i.e. the connect-the-dots and perceive-bond-order routines that we currently do with OpenBabel?

[pierrelb]

Not entirely certain. I haven't found methods that do that as yet, though there is a lot to RDKit that it still unexplored. It does a good job determining distances when given the connectivity, so I'd assume there would be methods to do the reverse - hence I'll keep looking.

[gmagoon]

There is a perceiveConnectivity() routine in MoleCoor: https://github.com/gmagoon/MoleCoor/blob/master/MolecularCoordinates.py#L53 . It is used as an alternative to the OpenBabel approach in connectivity-checking in RMG-Java. I'm not sure whether RDKit has this functionality.

[rwest]

Thanks @gmagoon. That's looks like it's equivalent to OpenBabel's ConnectTheDots method, which seems reasonably straight forward (Pythagoras with two atomic radii and a threshold). Did you ever find OpenBabel's PerceiveBondOrder method (that does all sorts of rule-based checks, electronegativity checks, valence checks, distance checks, etc.) to be useful or necessary (to figure out which are double bonds), or do you just do graph isomorphism without labelled edges (i.e. a bond is just a bond) to check you've found the right structure? If the latter, then we can probably do without OpenBabel.

@pierrelb, do you know if rdkit can read @connie's furans?

[gmagoon]

@rwest: no need for bond order perception because graph isomorphism is done via InChI, which only considers what is connected, not how it is connected (i.e. InChI only localizes one pair of electrons per bond)