Some SMILES from E.coli dataset can not be correctly processed.

[Phuangji]

Hello! When I predict kcat of some E.coli reactions, it says my SMILES are out of the range of atoms like this. For example, CC1(CC(=O)O)C2=Cc3[nH]c(c(CCC(=O)O)c3CC(=O)O)Cc3[nH]c(c(CC(=O)O)c3CCC(=O)O)C=C3N=C(C=C(N2)C1CCC(=O)O)C(C)(CC(=O)O)C3CCC(=O)O

Traceback (most recent call last):
  File "predict.py", line 82, in <module>
    pred = M( atoms_pad, atoms_mask, adjacencies_pad, amino_pad, amino_mask, batch_fps, inv_Temp, Temp )
  File "/torch/nn/modules/module.py", line 1194, in _call_impl
    return forward_call(*input, **kwargs)
  File "/DLTKcat/code/DLTKcat.py", line 161, in forward
    atoms_vector = self.comp_gat(atoms, adjacency)
  File "/DLTKcat/code/DLTKcat.py", line 102, in comp_gat
    atoms_vector = self.embedding_layer_atom(atoms)
  File "/torch/nn/modules/module.py", line 1194, in _call_impl
    return forward_call(*input, **kwargs)
  File "/torch/nn/modules/sparse.py", line 162, in forward
    self.norm_type, self.scale_grad_by_freq, self.sparse)
  File "/torch/nn/functional.py", line 2210, in embedding
    return torch.embedding(weight, input, padding_idx, scale_grad_by_freq, sparse)
IndexError: index out of range in self

I use CPU here. If use GPU there will be similar errors. Could you please tell me what's the problem with my operation? Looking forward to your reply. Thank you!

[SizheQiu]

I think the "[nH]" in the SMILES string is the issue. In the training dataset I used, all SMILES strings were canonical SMILES, and those with ions were all filtered out. Please make sure the SMILES string of the substrate was in canonical form.

[Phuangji]

Thank you for your reply. There are still some questions. First, I did convert my SMILES to canonical forms , and the code is as follows:

class MolClean(object):
    def __init__(self):
        self.normizer = MolStandardize.normalize.Normalizer()
        self.lfc = MolStandardize.fragment.LargestFragmentChooser()
        self.uc = MolStandardize.charge.Uncharger()

    def clean(self, smi):
        mol = Chem.MolFromSmiles(smi)
        if mol:
            mol = self.normizer.normalize(mol)
            mol = self.lfc.choose(mol)
            mol = self.uc.uncharge(mol)
            smi = Chem.MolToSmiles(mol,  isomericSmiles=False, canonical=True)
            return smi
        else:
            return None

Second, I can normally handle some SMILES that contain [nH], such as: O=c1[nH]c(=O)c2[nH]cnc2[nH]1 Also, the error SMILES does not have ions , it only contains a nitrogen heterocycle. So maybe your reply does not solve my problem. Thank you again!

[SizheQiu]

Ok, in this case, the most possible reason is that the substrate contains some molecular fingerprints that are not in my training data, and thus the substrate cannot be encoded as features.