Google Sumer of Code 2025: Preamble README

[singjc]

Welcome to the Diffusion Deconvolution DIA-MS Project

👋 Welcome Contributors!

Thank you for your interest in the project!

Note

We are currently in the community building phase (February 27th - March 24th), see the GSOC 2025 timeline. So there is no rush to jump straight into coding

Important

During the community building phase of Google Summer of Code (GSoC), we expect those interested to introduce themselves, submit their application to OpenMS and to Google and to get familiar with the codebase. See GSoC Contributors

🔍 Getting Started

• Visit our GSoC 2025 Wiki for recommended literature
• Browse Issues: Check our issue tracker for tasks labeled "GSoC 2025" to find entry points for contribution.
• Set Up Environment: Instructions for setting up the development environment are available in the REAMD.md

🚀 Project Overview

We're investigating the use of diffusion models to deconvolve complex, overlapping signals in DIA-MS data - similar to how generative AI models enhance images or separate speech from noise.

📚 Key Resources

• Code: Explore our jupyter notebooks and model implementations
• Data: We are currently using/generating from Searle, B. et. al. (2018) paper on chromatogram libraries. See the data availability section. Take a look at nbs/hela_2018.ipynb
  • We have raw data in sqMass format (sqlite db) in raw_data/
  • We have a few processed training data samples (parquet files) for narrow window DIA data here processed_data_for_training/ms_data_slices_narrow_.tar.xz
  • The original model was trained on a smaller dataset, that can be found here
  • We do have saved checkpoints for the DDIM(Unet1d) model, which can be found here
• Communication: Comment on issues or reach out to mentors with questions

🤝 How to Contribute

• Read the contribution guidelines in CONTRIBUTING.md
• Pick an issue that matches your interests and skills
• Comment on the issue to express interest
• Fork the repository and submit a pull request with your solution

We look forward to your contributions!

[dev-coder2]

Hi @singjc ,

I’m Dev kumar Acharya, a 4th-year student at IIT Bombay with experience in Python, PyTorch, and deep learning. I’m very excited about this project and its application of diffusion models to DIA-MS deconvolution. The idea of using MS1 as a guiding signal for deconvolving MS2 is fascinating, and I’d love to contribute to optimizing data loading, enhancing conditioning signals, and exploring transformer-based architectures.

[singjc]

Hi @singjc ,

I’m Dev kumar Acharya, a 4th-year student at IIT Bombay with experience in Python, PyTorch, and deep learning. I’m very excited about this project and its application of diffusion models to DIA-MS deconvolution. The idea of using MS1 as a guiding signal for deconvolving MS2 is fascinating, and I’d love to contribute to optimizing data loading, enhancing conditioning signals, and exploring transformer-based architectures.

Thank you for the interest in the project! Please see the wiki for literature on DIA-MS/MS data, and see the issues for the current potential tasks for contributing.

[dev-coder2]

@singjc can you share the previously trained state of our model.

[singjc]

@singjc can you share the previously trained state of our model.

I am trying to upload it to filetransfer.io, but it seems to take a long time. The file is ~13Gb...

[ShubhamChauhan22222]

@singjc can you provide me the "We do have saved checkpoints for the DDIM(Unet1d) model. If you would like to play around from a previously trained state, we can share this with you. However, the file is ~ 14Gb.". I want to work on static quantization of the pre-trained model.

[singjc]

@dev-coder2 @ShubhamChauhan22222 @CoderOMaster

I have a temporary link here to a pre-trained model

[CoderOMaster]

@singjc i am unable to open link and i am getting Error: Quick Share could not be found. Do not close the browser window before the transfer is complete.

[singjc]

@singjc i am unable to open link and i am getting Error: Quick Share could not be found. Do not close the browser window before the transfer is complete.

Can you try this one

[singjc]

@singjc i am unable to open link and i am getting Error: Quick Share could not be found. Do not close the browser window before the transfer is complete.

@CoderOMaster I've change the link to here, which seems to be working. @bhautik-pithadiya I think you were able to download from this link right?

[CoderOMaster]

@singjc i am able to download the file but unfortunately it provides very slow download speed. I will let it continue anyways but by any chance would it be possible for you to break this into multiple smaller .rar/.zip format and upload on something like gdrive/dropbox/onedrive etc?

[bhautik-pithadiya]

. @bhautik-pithadiya I think you were able to download from this link right?

No, the download speed was so slow that it got canceled.

any chance would it be possible for you to break this into multiple smaller .rar/.zip format and upload on something like gdrive/dropbox/onedrive etc?

@CoderOMaster That's a good idea. But would DVC be a better option for this? @singjc