[Rdkit-discuss] kekulize AllChem.CanonSmiles error and workaround

[Konrad K. <kon...@ic...>]

Hi,

I am having trouble canonicalizing smiles with ambiguous heteroaromatic tautomers such as imidazole. For example:

>>> from rdkit import Chem
>>> from rdkit.Chem import AllChem
>>> smiles = ‘n1cncc1'
>>> AllChem.CanonSmiles(smiles)
[21:42:52] Can't kekulize mol.  Unkekulized atoms: 0 1 2 3 4

As a workaround, one can first canonicalize with Open Babel pybel to remove the ambiguity and then canonicalize with RDKit:

>>> import pybel
>>> pybel.readstring("smi", "n1cncc1").write("can")
'c1ncc[nH]1\t\n'
>>> AllChem.CanonSmiles('c1ncc[nH]1\t\n')
'c1c[nH]cn1’

or in one line:

>>> AllChem.CanonSmiles(pybel.readstring("smi", "n1cncc1").write("can"))
'c1c[nH]cn1'

It would be nice if RDKit could do this without the assistance of pybel.



This problem arose when implementing the algorithm described in the following paper:

Hall RJ, Murray CW, Verdonk ML. The Fragment Network: A Chemistry Recommendation Engine Built Using a Graph Database. J Med Chem. 2017; 60(14):6440-50. PMID: 28712298, doi: 10.1021/acs.jmedchem.7b00809

Details of the algorithm are contained in supporting information:
http://pubs.acs.org/doi/suppl/10.1021/acs.jmedchem.7b00809/suppl_file/jm7b00809_si_001.pdf <http://pubs.acs.org/doi/suppl/10.1021/acs.jmedchem.7b00809/suppl_file/jm7b00809_si_001.pdf>

The algorithm fragments the molecule at acyclic bonds connected to rings and it is necessary to canonicalize both the parent and child fragments. The algorithm is recursive and fortunately the smiles can be recursively processed by AllChem.CanonSmiles after it has been disambiguated:

>>> AllChem.CanonSmiles('c1c[nH]cn1')
'c1c[nH]cn1’

I eventually plan to donate the RDKit Fragment Network script to the community after testing and optimization.

Best,

Konrad