How to Fine-Tune with PDBs Outside the Preprocessed Dataset?

[rcossio]

The training documentation mentions that fine-tuning requires a list of PDB files and the preprocessed wwwPDB dataset (~1TB).

I would like to fine-tune the model using structures that are not included in this preprocessed dataset. Could you provide guidance on how to include external PDB files in the fine-tuning process?
I’m considering defining custom IDs, but should the structures be preprocessed to be added to the dataset?

Thank you in advance for your help!

[zhangyuxuann]

@rcossio We consider release the data pipeline in the feature. This matter needs to go through some processes. If there is any update, it will be synchronized here.

[Linmj-Judy]

Hi, yuanxuan, hope this email finds you well.

Your proposed algorithm has captured my keen research interest.

Following up on the previous inquiry about the data processing pipeline, I would also like to express my strong interest in your work.

The algorithm looks promising, and I am eager to see its implementation details.

Thank you for your time and consideration.

Warm regards,
Mujie Lin

[xinshi-chen]

Hey everyone, the data pipeline was released last week! You can find more details here:
@rcossio @Linmj-Judy

https://github.com/bytedance/Protenix/blob/main/docs/prepare_training_data.md