Exploring the Dark Matter in a Mass Spectrometry Proteomics Dataset

[edeutsch]

Title

Exploring the Dark Matter in a Mass Spectrometry Proteomics Dataset

Abstract

Typical analyses of DDA mass spectrometry data still only identify a small subset of the MS2 spectra above accepted quality thresholds, although there is wide variation among different datasets. The spectra with identifications below threshold are sometimes referred to as the “Dark Matter” of mass spectrometry proteomics experiments, and are generally discarded from further analysis. We hypothesize that many and perhaps most of these dark matter spectra can be identified with concerted effort by the community and we could create a gold standard set of identifications. They can likely be explained by chimeric spectra, insufficiently scoring but correct identifications, uncommon mass modifications, incorrect precursor m/zs, unconsidered contaminants, and combinations thereof. We will build a platform to enable the participants to throw various analysis tools, including sequence search, spectral library search, de novo search, open modification search, AI tools, and manual evaluation at one dataset and synthesize the results into a shared gold standard identification list. This will result in greater understanding of the dark matter problem and provide a high-quality benchmark dataset for better evaluation of software tools and FDR calculations.

Project Plan

These are the tasks we imagine:

• Agree on the target dataset (perhaps an immunopeptidomics dataset)
• Design and build (or build on) a platform for integrating results from various identification “submissions” of results (perhaps using ProteoBench infrastructure)
• Run various search engine/analysis tools on the dataset and upload results
• Design and build a mechanism for human review of tool disagreements using Universal Spectrum Identifiers
• Design and build tools to summarize the progress in identifying all spectra
• Design and build a mechanism for classifying unidentified spectra into categories (low quality and not identifiable, high quality and should be identifiable, not a peptide, confidently identified, etc.)

During the hackathon we will therefore do:

1. Select the target dataset
2. Set up the infrastructure to track the answers and final result for every MS2 scan in the dataset
3. Run many analysis tools on the dataset and upload the results
4. Achieve a consensus on as many spectra as possible
5. Categorize all identified and unidentified spectra
6. Develop a plan for how to finish and write a manuscript describing the project and what we learned from the effort

After the hackathon we will seek expertise and contributions from experts in the field of spectrum interpretation and write a manuscript describing the project and what we learned about proteomics dark matter.

Technical Details

Programming language: Python
Will build on existing software: ProteoBench, existing search engines, USIs
These datasets will be used: One immunopeptidomics dataset from ProteomeExchange and already reanalyzed in PeptideAtlas, but still TBD

Contact information

Eric Deutsch
Institute for Systems Biology
Seattle WA USA
edeutsch@systemsbiology.org