Sporadic errors when generating RDKit conformers that match training (crystal) conformers

[amorehead]

Hello. Thank you all for your great work on DiffDock and similar efforts! I was curious if anyone else has experienced sporadic (i.e., periodic) Runtime Errors in Python when executing SciPy's differential_evolution function on an RDKit conformer generated to match a crystal conformer in DiffDock's validation dataset.

https://github.com/gcorso/DiffDock/blob/ac5612da4fae0aa6f63faff3e1586c474ee307fb/datasets/process_mols.py#L301

Specifically, the error reads as RuntimeError: No sub-structure match found between the probe and query mol. It seems like SciPy in this case failed to find a match between the generated conformer and the ground-truth conformer in the validation dataset when it tried to optimize the generated conformer's rotatable bonds. Has anyone else encountered an issue like this, and if so, how did you resolve it?

Thanks!