molecule.fromSMILES(CC1=CC=CO1) gives "Unable to determine atom type for atom O."

[rwest]

Try 2-methyl furan: http://rmg.mit.edu/adjacencylist/CC1=CC=CO1

Gives Unable to determine atom type for atom O.

[stroiano]

From the SMILES 'CC1=CC=CO1' (or, equivalently, 'Cc1ccco1'), OpenBabel generates an invalid aromatic structure:

Trying

mol = Molecule().fromAdjacencyList('
    1 C 0 {2,D} {5,S}
    2 C 0 {1,D} {3,S}
    3 O 0 {2,S} {4,S}
    4 C 0 {3,S} {5,D} {6,S}
    5 C 0 {1,S} {4,D}
    6 C 0 {4,S}
')

generates the proper structure:

but then trying:

mol2 = Molecule().fromSMILES(mol.toSMILES())

gives the same atom type error.

So, perhaps there's some way for OpenBabel to recognize and fix such aromaticity problems? Adding obmol.Kekulize() before building the Molecule object didn't seem to help and would probably lead to problems with multi-ring structures anyway.

[rwest]

@nickvandewiele - you're the aromaticity expert around here - what do you think?

[stroiano]

Here (in lines 64-89) is a (very) rough modification to Molecule.fromOBMol that seems to fix this issue. It looks insane, but it works.

Perhaps it would be a good idea here to implement a robust Kekulization method before attempting to assign atom types to the molecule.

Of course, the other alternative is to simply define Ob and Obf atom types...

[rwest]

Hmm. Aromaticity is tricky stuff! http://www.dalkescientific.com/writings/diary/archive/2007/11/30/opensmiles_and_aromaticity.html

But some way to Kekulize would probably help us. If we can off-load it to an external library like OpenBabel then great, because they'll have already put in lots of thought and debugging.

Anyone know what RMG-Java thinks of these structures? What "type" is that O atom?

[nickvandewiele]

ok, here are my thoughts and comments on this:

• OB's SMILES parser always performs an aromaticity perception step for its own atom type perception algorithms. RMG does not (in RMG-Py at least).
• we don't have an aromatic oxygen atom type in RMG. adding one will require additions to thermochemistry databases and kinetics databases
• if you only kekulize oxygen in case of furan, you will get into trouble with thermochemistry accuracy. because Cb benson group contributions already include the stabilization effect of the aromatic ring. this means that no additional ring "strain" for benzene needs to be added on top of the 6 Cb atoms. each Cb has 1 6th of the ring stabilization in benzene. In the case of furan, you will add 4 Cb's and one Os atom. 4 6ths of the benzene ring stabilization is probably not a good guess for furan ring stabilization. you need to completely kekulize furan, add Cds, Os contributions, and add a furan ring correction
• In RMG-Java I (likely we) only looked at at pure hydrocarbon aromaticity issues. haven't tested it though.
• I agree in externalization of kekulization. having a particular molecule in your head while writing the algorithm is often a bad idea in the long run. moreover, as aromaticity perception is already very tricky in eliminating false positives/negatives. the reverse process is probably even more complex for the correction assignment of D, S bonds.
• I agree that we need to go for more atom types, like Ob simply because oxygen in furan does not behave as oxygen in butanol. more atom types means more adequate matches but more required data to fill up the tree. all chemo-informatics packages have this tendency to add more and more atom types btw. their logic is to better let it crash loudly than to silently assign false numbers.
• (De)-aromatization is still an issue in chemo-informatics libraries, even after say 20 years. partially because aromaticity is not well defined. so don't expect to fix this weekend. :)

[nickvandewiele]

furan thermo with RMG-Java on master (commit ca. july 2012) through the ThermoDataEstimator with adjacency list gives the following thermochemistry:

-4.3 74.95 20.79 27.14 32.51 36.99 44.07 49.03 56.08

all carbons are Cds, oxygen Os. furan ring correction (DH°f(298K) 6.89 kcal/mol) is applied.

NIST webbook reports DH°f(298K) of -6.6 kcal/mol (pretty old, 1991) so 2.3 kcal off. not that bad, i'd say.

[nickvandewiele]

btw: is the Molecule().fromSMILES() a vital method in RMG-Py?

otherwise i'd propose a very clever patch: "do not use SMILES parsers in case of tricky aromatic-like compounds"

[rwest]

I think Molecule().fromSMILES() is vital for various parts of the website.