Chai prediction with ligands

[BJHardy]

Hi!

I've noticed two bugs relating to Chai1 prediction of ligands when running the pipeline.

1. Ligands are incorrectly added as 'protein' entries into the Chai1 input fasta files.
2. If backbones are generated around two ligands of the same name, then they will both be incorrectly given the same residue number in the RFdiffusion2 output backbone. This doesn't affect the LigandMPNN step, but it means that only one copy of the ligand is added to the Chai1 input fasta file. This means the predicted structures only have one of two ligands, throwing off the metrics step.

Happy to share more info if needed. Thanks!

[rclune]

@dtischer @w-ahern @r-krishna

Do any of you know of a fix or work-around for this?