UnlabledSmiles errors on forward pass when passing valid single atom smiles

[sudoPete]

There appears to be an edge case in the UnlabeledSmiles data loader that fails to initialize the edges attribute for a single heavy atom smiles string. This causes a key error when trying to access the edge features.

      data = UnlabeledSMILES(  
             ['C'],  
             node_featurizer=CanonicalAtomFeaturizer(),  
             edge_featurizer=CanonicalBondFeaturizer(),  
             )  
     smiles, graphs = map(list, zip(*data))
     bg = dgl.batch(graphs)
     bg.set_n_initializer(dgl.init.zero_initializer)
     bg.set_e_initializer(dgl.init.zero_initializer)
     node_feats = bg.ndata.pop('h')
     edge_feats = bg.edata.pop('e')

_collections_abc.py", line 795, in pop value = self[key]
dgl/view.py", line 181, in getitem
return self._graph._get_e_repr(self._etid, self._edges)[key]
KeyError: 'e'

[mufeili]

This is because that you have a single atom and there are no chemical bonds. As a result, the molecular graph has no edges, hence no edge features. For a workaround, try the solution below, which adds self loops.

from dgllife.data import UnlabeledSMILES
from dgllife.utils import mol_to_bigraph, CanonicalAtomFeaturizer, CanonicalBondFeaturizer
from functools import partial

data = UnlabeledSMILES(['C'], mol_to_graph=partial(mol_to_bigraph, add_self_loop=True), 
                                            node_featurizer=CanonicalAtomFeaturizer(), 
                                            edge_featurizer=CanonicalBondFeaturizer(self_loop=True))