Re: [Rdkit-discuss] Question on chirality

[Jan H. J. <ja...@bi...>]

Hi Navid,

I am not familiar with the paper you mention, but I believe that the 
problem is caused by non-isomeric input SMILES.

Below is an Alanine read in from molfile, with coordinates. It has a 
chiral center with "S" configuration. When you output it as non-isomeric 
SMILES and read it back in, the chiral information is lost because the 
molecule no longer has a conformation:

     >>> mol = Chem.MolFromMolBlock("""
    ...   BIOCHEMF09131911262D
    ...
    ...   7  6  0  0  1  0  0  0  0  0999 V2000
    ...     0.0000    0.0000    0.0000 N   0  0  0  0  0  0  0 0  0  0  0  0
    ...     0.7145    0.4125    0.0000 C   0  0  0  0  0  0  0 0  0  0  0  0
    ...     1.4290    0.0000    0.0000 C   0  0  0  0  0  0  0 0  0  0  0  0
    ...     1.4209   -0.8208    0.0000 O   0  0  0  0  0  0  0 0  0  0  0  0
    ...     0.7084    1.2417    0.0000 C   0  0  0  0  0  0  0 0  0  0  0  0
    ...    -1.0000    0.0000    0.0000 H   0  0  0  0  0  0  0 0  0  0  0  0
    ...     2.4290    0.0000    0.0000 O   0  0  0  0  0  0  0 0  0  0  0  0
    ...   2  3  1  0  0  0  0
    ...   1  2  1  0  0  0  0
    ...   3  4  2  0  0  0  0
    ...   2  5  1  1  0  0  0
    ...   1  6  1  0  0  0  0
    ...   3  7  1  0  0  0  0
    ... M  END
    ... """)
     >>> Chem.AssignAtomChiralTagsFromStructure(mol)
     >>> Chem.FindMolChiralCenters(mol)
    [(1, 'S')]
     >>> Chem.MolToSmiles(mol)
    'CC(N)C(=O)O'
     >>> mol = Chem.MolFromSmiles("CC(N)C(=O)O")
     >>> Chem.AssignAtomChiralTagsFromStructure(mol)
     >>> Chem.FindMolChiralCenters(mol)
    []
     >>>


You can generate a conformation that produces chiral information by 3D 
embedding the molecule.

     >>> from rdkit.Chem import AllChem
     >>> AllChem.EmbedMolecule(mol)
    0
     >>> Chem.AssignAtomChiralTagsFromStructure(mol)
     >>> Chem.FindMolChiralCenters(mol)
    [(1, 'S')]
     >>>


Another way would be if you can get isomeric SMILES as input. Then the 
chiral information is right there.

     >>> Chem.MolToSmiles(mol, isomericSmiles = True)
    'C[C*@*H](N)C(=O)O'
     >>> mol = Chem.MolFromSmiles("C[C@H](N)C(=O)O")
     >>> Chem.FindMolChiralCenters(mol)
    [(1, 'S')]
     >>>


Cheers
-- Jan Holst Jensen


On 2019-09-12 04:44, Navid Shervani-Tabar wrote:
> Hello,
>
> In the paper: "Graph Networks as a Universal Machine Learning
> Framework for Molecules and Crystals", authors introduce chirality as 
> an atom feature input to analyze QM9 dataset. I was trying to recreate 
> this atom feature as following
>
> > Chirality: (categorical) R, S, or not a Chiral center (one-hot encoded).
>
> The code I used is:
>
>     from chainer_chemistry import datasets
>     from chainer_chemistry.dataset.preprocessors.ggnn_preprocessor 
> import GGNNPreprocessor
>     from rdkit import Chem
>     import numpy as np
>
>     dataset, dataset_smiles = datasets.get_qm9(GGNNPreprocessor(), 
> return_smiles=True)
>
>     for i in range(len(dataset_smiles)):
>         mol = Chem.MolFromSmiles(dataset_smiles[i])
>         Chem.AssignAtomChiralTagsFromStructure(mol)
>         chiral_cc = Chem.FindMolChiralCenters(mol)
>
>         if not len(chiral_cc) == 0:
>             print(chiral_cc)
>
> The output shows no Chiral centers for this dataset. When I use 
> `includeUnassigned=True`, code gives a list of tuples, but instead of 
> "R/S", I get "?". I was wondering if there is a mistake in my 
> implementation. If this is expected, any thoughts on how chirality was 
> assigned in the above paper? Thanks.
>
> Sincerely,
> Navid