Using ProteinMPNN *without* oxygens?

Hi,
I noticed that if I run ProteinMPNN on backbones *without* the oxygens, the model does *not* break; however, the designs look very different with vs. without the addition of oxygens.
My question is: should I expect the designs made *without* oxygens to be non-sensical? if this is true, then I wonder why the model doesn't throw an error instead.