With-out being familiar with that DB and data set and my knowledge of that being a bit rusty (i.e. 10 years old) and trying to interpret what you're doing:
I would say you can construct a "formal" network of atoms and their bonds from the protein's sequence. This completely ignores secondary and tertiary structure, where bonds and van der Waals forces play a role.
Assuming that the data you're downloading is X-ray crystallography or some such, mapping the protein into that and deriving bonds and near bonds and such, is AFAIK an open research question. You may have better luck in a computational biology or computational chemistry related forum to get updates on this.
Please let me know, if I misinterpreted anything. I'll then delete this answer.