Request for data acquisition and preprocessing pipeline for MassSpecGym on Iceberg

[WinterWei22]

Hi! Thanks for your great job!

I’d like to process the MassSpecGym dataset on Iceberg, but I couldn’t find the data acquisition and preprocessing pipeline. Could you please provide the relevant code or point me to the appropriate scripts/documentation?

Many thanks in advance for your time and support. I sincerely appreciate your work on this project.

With best regards, @rogerwwww @connorcoley

[mrunalimanj]

Hi, thank you for your interest! The codebase should be reflected to support training on the MassSpecGym dataset. The main difference in architectural implementation is the inclusion of an instrument embedding, since both Orbitrap and QToF data are present in the dataset.

You can find the weights readily available at the top of the README, and all the scripts in run_scripts, alongside the relevant configs (*_msg_spec.yaml) should be available. If you find any errors or run into difficulty with inference, training, or other steps, please feel free to follow up below.

[WinterWei22]

Thanks so much for the support—really appreciate it!
I noticed the Dropbox you shared only includes the retrieval candidates and pretrained iceberg weights. Would you mind sharing the full dataset as well? Having the complete data would make it much easier for us to reproduce the results.
Thanks again for your help! :)