Can I access the 22-class model including endosome prediction?

[luckyvivi]

Hi Protgps team,

Thank you for sharing this great tool!

I’m particularly interested in predicting endosome localization. I ran the Predict.ipynb notebook from your repository, which works well. However, the model it uses only outputs 12 compartments, and endosome is not among them.

I noticed in your source code (e.g., ProteinCompartmentsUniprotCombined dataset) that there’s a configuration for 22 compartments, including endosome. May I ask:
• Have you released a pretrained model checkpoint trained on the 22-compartment dataset (e.g., protein_compartment_uniprot_combined)?
• If so, where can I download it?
• What’s the recommended way to use it in the existing pipeline (e.g., do you have a corresponding .args and .ckpt)?

Thanks again for your amazing work, and I really appreciate your help!

Best regards,
Vivian.

[pgmikhael]

Hi Vivian!

We only implemented the dataset but do not have a model for the other compartments. One could train a model to predict them in a similar fashion. The only snapshot files we have are for the published 12-compartment model.

Thanks for reaching out!