Training with fold exclusion

[Teohoho]

Hello! I've been using Openfold for a while, mostly for inference, but I would like to try training it for a specific type of fold. From your paper I understand you've done something similar where you trained it by using exclusively Mostly Alpha/Mostly Beta structures to gauge how well it'd predict unknown fold space. I was wondering, if I want to do something similar, more specifically I have a list of mmCIFs I want excluded from training, would it be sufficient to just remove these files from the directory that mmcif_dir argument points to in scripts/generate_mmcif_cache.py , scripts/generate_chain_data_cache.py and train_openfold.py?

Thank you!