Input direct from PDB file fails

Hi there,

Thanks for the early release of the model, it's great to look at and play around with!

I am having issues with running predictions with ligands directly from a PDB file. It appears to be an issue with finding chiral centres with rdkit. Both the examples modelforge/tests/data/example_from_pdb_with_inter_chain_bond.pdb and modelforge/tests/data/example_pdb_with_clashing_ligand_name.pdb fail when loaded directly with rf3 fold inputs='modelforge/tests/data/example_from_pdb_with_inter_chain_bond.pdb'

Symptoms: the prediction hangs for a long time at WARNING:atomworks.ml.transforms.rdkit_utils:Initial conformer generation based on distance geometry failed. Successful: 0. Falling back to generating a conformer starting from random coordinates. example_pdb_with_clashing_ligand_name hangs for approx 2 hours. example_pdb_with_inter_chain_bond hangs for over 12 hours.

Afterwards it fails to find chiral centres with rdkit, and the run fails.

PDB files with no ligand run successfully. Log files attached for both examples.

Thanks.

example_pdb_with_inter_chain_bond.log
example_pdb_with_clashing_ligand_name.log

Reproduce with the rf3 fold command and the two inputs in modelforge/tests/data/example_from_pdb_with_inter_chain_bond.pdb and modelforge/tests/data/example_pdb_with_clashing_ligand_name.pdb. Start by examining the attached logs and the RDKit chiral-centre and conformer-generation failure. Done means ligand-containing PDB inputs complete prediction without hanging or failing to find chiral centres.