Hi Ling, > On Apr 2, 2021, at 16:23, Ling Chan <lingtrek...@gmail.com> wrote: > > Thank you Francois, I took a look at your code and borrowed parts of it to > rejoin two molecules. It seems like my problem is solved. I eventually > arrived at something like example 4 in > https://www.programcreek.com/python/example/123334/rdkit.Chem.CombineMols > (which I discovered a bit late). > > Still, I am not sure if the code is safe. In particular, I wonder if the > following conditions are always valid. > • Chem.CombineMols simply concatenates the atomic indices from the > input molecules. > • The Chem.EditableMol constructor preserves atom ordering from the > input. > • RemoveAtom in EditableMol results in all indices above the deleted to > decrease by one, i.e. atom ordering is preserved.
I've found that it's very hard to work with molecular graphs and preserve stereochemistry. Consider F/C=C/Cl breaking on the first bond, and the code I pointed you to. FragmentOnBonds() using '9' as the labels gives: [9*]/C=C/Cl.[9*]F My "smiles_weld" code converts that to: CC\%99=C/Cl.F%99 which can be re-canonicalized to the original: F/C=C/Cl . Or, with F[C@H](Cl)Br again, breaking on the first bond. FragmentOnBonds() gives [9*]F.[9*][C@H](Cl)Br smiles_weld converts that to F%99.[C@@H]%99(Cl)Br which is re-canonicalized as F[C@H](Cl)Br Handling this correctly in the molecule API requires paying careful attention to the bond direction, and bond attachment order around the atom, which changes with RemoveAtom() calls. I didn't see stereochemistry support in Francois's "bind_molecules()" nor in the connect_mols() at https://github.com/molecularsets/moses/blob/master/moses/baselines/combinatorial.py (one of the examples from the programcreek.com link you gave). If you don't need to support or preserve stereochemistry, then of course there's no problem. Cheers, Andrew da...@dalkescientific.com _______________________________________________ Rdkit-discuss mailing list Rdkit-discuss@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/rdkit-discuss
