Research plan for the spectrafuse tool

[ypriverol]

Spectral Clustering of quantms Data and Spectral Library Generation

Introduction

quantms, a cutting-edge workflow, has reanalyzed an extensive dataset of almost 1 billion MS/MS (Mass Spectrometry/Mass Spectrometry) scans, comprising nearly 100 million PSMs (Peptide-Spectrum Matches) derived from various tissues, cell lines, and diseases. In light of this vast wealth of data, our project aims to apply spectral clustering techniques to organize this data and construct spectral libraries.

Project Goals

• Development of an Incremental Clustering Algorithm: The primary objective of this project is to design a novel algorithm for spectral clustering that leverages an incremental clustering approach. This approach will allow us to efficiently organize the vast volume of MS/MS data generated by quantms. The algorithm and workflow should be implemented in nextflow using the popular tool MaRACluster (see appendix).

  • Remove the mraracluster as parameter #5

• Creation of New Spectral Libraries: We will generate updated spectral libraries based on the comprehensive quantms dataset. These libraries will serve as invaluable resources for researchers in the field of mass spectrometry, enabling them to identify peptides and their corresponding spectra.

• Development of a Command-Line Search Tool: A crucial component of our project is the creation of a command-line tool. This tool will enable users to search for specific peptide sequences or spectra within the spectral libraries using the innovative algorithm, as described in our previous publication [https://www.sciencedirect.com/science/article/pii/S1874391920304383]. This user-friendly tool will significantly enhance the accessibility and usability of the spectral libraries.

• Application to Proteogenomics and PTMs: Once we have successfully developed the algorithm and infrastructure, we will expand our scope to apply these tools to proteogenomics datasets and the analysis of Post-Translational Modifications (PTMs). This extension will provide researchers with a comprehensive platform for studying protein-related phenomena beyond traditional peptide identification.

Expected Impact

The successful completion of this project will have significant implications for the field of mass spectrometry and proteomics. Researchers and scientists will benefit from more efficient data organization and advanced search capabilities, which will enhance their ability to extract meaningful insights from large-scale MS/MS datasets. Additionally, the development of spectral libraries and their application to proteogenomics and PTM analysis will contribute to a deeper understanding of complex biological systems and disease mechanisms.

Conclusion

We are excited to embark on this project to harness the power of spectral clustering and deep learning algorithms to revolutionize the analysis of MS/MS data. With your support, we can make significant strides in advancing the field of mass spectrometry and proteomics. We look forward to the opportunity to collaborate and bring our innovative approach to fruition.

Appendix:

Because the amount of data and volume is too big, we aim to implement an incremental clustering strategy. Instead of using spark as previously, we can implement an interactive/incremental approach that for every cluster iteration we keep the most relevant/high quality PSMs by cluster, the number can be tested from 10 to 20.

Reference: https://github.com/bigbio/spectrafuse/blob/main/docs/algorithm.png

Workflow Notes

Tool mgf-converter: We should take every project here and convert them into MGF files. Note: We need to think carefully here because some of the projects have more than 30 M PSMs, and we should not create one MGF file per project because it will be too big. I recommend here to use at the beginning a batch size effect, by default 1M spectra per MGF File.

We will be clustering together only spectra from the same Instrument, Species and charge state. Then probably would be good in this conversion to create the files based on those three metrics.

MAraCluster: With the MGF files, you should run MAraCluster tool and perform clustering. Remember that every clustering run should only contain the MGFs for the spectra that are from the same Species, charge and Instrument.

Even if we cluster together only spectra from the same species, charge and instrument we may find that for some combinations of those three variables we have hundreds of millions of spectra, which probably will make

Maracluster is not to scale properly. If that is the case we may explore an incremental approach where we cluster for example 10M spectra, then we can select for that batch the spectra based on:

For clusters with more than 20 spectra, we selected the best 5 spectra with better signal-to-noise ratio + better ID score (qvalue); or we can just generate the for every cluster the consensus spectra and use that in the next iteration. The problem with this approach is that we need to be sure the quality of the cluster does not deteriorate over cluster iterations.

Library converter:
- After all the clustering is done we should have a folder with the corresponding structure:
- Species Name
- Instrument Name
Libraries1-charge2
Library2-charge3