How to correctly parse the SMILES of the PubChem dataset?

[Raimondas Galvelis]

The SMILES of the PubChem dataset are generated with OpenFF-Toolkit (https://github.com/openmm/spice-dataset/blob/main/pubchem/createPubchem.py). So, Molecule from OpenFF-Toolkit should be able read them correctly, but this isn't a case.

Get a SMILES:

import h5py

h5 = h5py.File('pubchem/pubchem-1-2500.hdf5')
smiles = h5['103914790']['smiles'][0]
print(smiles)

b'[N:1]1=[C:2]2[N:3]([C:5]([H:17])([H:18])[C:4]1([H:15])[H:16])[C:12]1([H:30])[C:8]([H:23])([H:24])[C:13]3([H:31])[C:6]([H:19])([H:20])[C:11]2([H:29])[C:7]([H:21])([H:22])[C:14]([H:32])([C:9]1([H:25])[H:26])[C:10]3([H:27])[H:28]'

Parse the SMILES and print elements:

from openff.toolkit.topology import Molecule

mol = Molecule.from_smiles(smiles, hydrogens_are_explicit=True, allow_undefined_stereo=True)
print([atom.element.symbol for atom in mol.atoms])

['N', 'C', 'N', 'C', 'H', 'H', 'C', 'H', 'H', 'C', 'H', 'C', 'H', 'H', 'C', 'H', 'C', 'H', 'H', 'C', 'H', 'C', 'H', 'H', 'C', 'H', 'C', 'H', 'H', 'C', 'H', 'H']

Despite the SMILES contains the explicit hydrogen and atom indices, the order of atom doesn't match, e.g. the 5th atom in the SMILE is C, but in the molecule it is H.

[Pavan Behara]

Hi @raimis , please use Molecule.from_mapped_smiles() which retains the atom mapping.

mol = Molecule.from_mapped_smiles(smiles, allow_undefined_stereo=True)

[Raimondas Galvelis]

@pavankum thanks! I haven't noticed that in the documentation.

[Pavan Behara]

Thank you for the feedback, we will make sure to update the documentation.