Different parents for same molecule with ring substituents

[conrad-stork]

Hi CSP Team (@greglandrum @eloyfelix @apbento),

first of all, a great thanks to your team for providing such a pipeline! I think this is a good step into the right direction to solve the mess in virtual libraries! I hope you will maintain and further develop this project. It is very useful!

I am not sure if I do something wrong, but for this little example with 3 same molecules, only two (the first and the third) get matched by the pipeline:

import chembl_structure_pipeline as csp
from rdkit.Chem import MolFromSmiles, MolToSmiles

def main():
    smis = [
            'c1c(C)c(C)ccc1',
            'c1cc(C)c(C)cc1',
            'c1ccc(C)c(C)c1']

    print('Input SMILES:')
    for smi in smis:
        print(smi)

    mols = [MolFromSmiles(smi) for smi in smis]
    std_mols = [csp.standardize_mol(mol) for mol in mols]
    par_mols = [csp.get_parent_mol(s_mol)[0] for s_mol in std_mols]
    par_smis = [MolToSmiles(p_mol) for p_mol in par_mols]

    print('Parent SMILES:')
    for p_smi in par_smis:
        print(p_smi)

if __name__ == '__main__':
    main()

Perhaps you can provide me some feedback to this example? I have several other examples that get not matched.

Best Conrad

[lnrtk]

Hi Conrad,

without looking deeper into the problem in this case a quick solution would be to canonicalize the smiles, e.g. via

par_smis = [MolToSmiles(MolFromSmiles(MolToSmiles(p_mol))) for p_mol in par_mols]

or something like:

mols = [MolFromSmiles(s) for s in smis]
mols = [MolToSmiles(m) for m in mols]
mols = [MolFromSmiles(s) for s in mols]

But I'm not sure if this should be the expected behaviour. Does this fix apply to your other problems or did you find a better way?

Best regards, Lennart

[greglandrum]

@conrad-stork : the problem here is that the molecules which come back from get_parent_mol() are not fully sanitized. If you sanitize the molecules before generating the SMILES things should work:

In [9]: for m in par_mols:
   ...:     Chem.SanitizeMol(m)
   ...:

In [10]: par_smis = [Chem.MolToSmiles(p_mol) for p_mol in par_mols]

In [11]: par_smis
Out[11]: ['Cc1ccccc1C', 'Cc1ccccc1C', 'Cc1ccccc1C']

@eloyfelix : we should think about adding a call to SanitizeMol() to the pipeline functions which return molecules.

[conrad-stork]

Hi,

thanks to @lnrtk and @greglandrum for the replies.

Both solutions (from Inrtk and greglandrum) are working. I will prefer the one from greglandrum.

From my point of view the issue is solved so I will close it. If you want to add the SanitizeMol greglandrum suggested, you can reopen it again. I would vote for it because I think it would make completely sense to include the "SanitizeMol feature".

Best Conrad

[eloyfelix]

Hi all,

sorry for the silence on this repo. We've been latelly quite involved on infrastructure works at EBI and we are also quite busy on the data side of things. I just want to reassure you that this is and will still be a core piece (hence mantained) of the ChEMBL database.

Thank you @greglandrum for keeping an eye on this. Yes, I think that it makes sense and I think that we could also re-evaluate if using RDKit's sanitization could also be an option for us.

There are few issues that we'd like to sort for ChEMBL31 (v30 to be released in ~1 month) and we'll tackle all of them at once since testing changes on the standardiser on the whole ChEMBL database usually takes us a bit of time.

[greglandrum]

@eloyfelix, just to confirm: the ChEMBL team is just using the _molblock() forms of the functions in your pipeline, right? So if we decided to change some of the behavior of the _mol() functions it wouldn't affect you?

[eloyfelix]

yes, @greglandrum, you're right. We are only using standardizer.get_parent_molblock and standardizer.standardize_molblock in our ingestion pipeline so changing the _mol() functions should be OK for us.