Jin Su

The original files are ``.pdb`` files. We just extracted such information and corresponding labels to generate a unified lmdb for data loading.

Hi, I think the way you mentioned that concatenating the sequences of different chains into one and feed it into SaProt is feasible. Besides, you may also obtain the embeddings...

We split a multi-chain protein into single-chain proteins and then adopted the same training strategy :)

Hi, It's good to include more metrics to comprehensively evaluate the performance of different models! Since the results were recorded around 2 years ago, we didn't save all model checkpoints...

Hi, In my opinion, you could formulate your senario as a classification task where label 1 indicates two proteins can bind and label 0 indicates they cannot bind. Then you...

Hi, thank you for your interest in our work! These three parts are not aligned directly. They are different databases for retrieval purpose. For example, If you have a function...

Hi, We have already released the pre-training code. You could refer to https://github.com/westlake-repl/SaProt/issues/24.

Hi, It looks like the node you used was busy at that time. Maybe it was executing an earlier search?

This is because the system cannot find the config file of protein encoder. Did you download the model weights correctly following https://github.com/westlake-repl/ProTrek?tab=readme-ov-file#Download-model-weights ?

Hi, Thank you for your feedback to make our document more clear! To avoid confusion, we have removed ``load_protein_pretrained`` and ``load_text_pretrained`` in the example. We keep these arguments for debugging...