Adding Charged Fragment Types to MS2 Model Weights

[mo-sameh]

This PR introduces a small but useful modification to the MS2 model interface and ModelMS2Bert,by including charged fragment types as an attribute within the model weights. With this change, charged fragment types are saved and loaded from disk, making it easier to adjust fragment types for prediction and training.

Main Benefits

• More flexibility in training and inference
  • Supports partial weight loading to continue training with different fragment types (as suggested in PR allow partial loading for pre trained ms2 models #226).
  • Allows prediction with any subset of supported fragment types, including masking modloss fragments—removing the need for a separate mask_modloss argument.
• Better handling of unsupported fragment types
  • Now raises clear and interpretable errors when an invalid fragment type is requested.

I've added a notebook (adapt_charged_fragtypes.ipynb) that walks through different use cases, including compatibility with older weight formats. It would be great if you could take a look and let me know if anything is missing.

This update modifies ModelMS2Bert, but if this approach looks good, I’d be happy to extend it to other architectures in a follow-up PR.

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB