Metabolomics most wanted: hunting the spectral unknowns of GNPS

[Janne98]

Title

Metabolomics most wanted: hunting the spectral unknowns of GNPS

Abstract

In untargeted metabolomics, tandem mass spectrometry (MS/MS) is a high-throughput technique for characterizing complex mixtures. However, only a small fraction of the resulting spectra can be annotated because reference spectral libraries cover just a sliver of chemical space. The remaining "dark" spectra represent an enormous, largely unexplored reservoir of chemical information.

Here we process the entire GNPS repository with Everything Bagel and Falcon to produce a deduplicated, repository-scale set of consensus MS/MS spectra suitable for large-scale annotation of chemicals and diverse ion species. Using spectral library searching, we will first annotate this set and reconstruct how annotation coverage has grown year over year, quantifying the impact of expanding public libraries on the (re)discovery of previously detected molecules. For unmatched spectra, we will provide putative annotations using cross-modal database retrieval accompanied with confidence scores.

We nonetheless hypothesize that a large fraction of spectra will remain unannotated, and the central aim of this project is to explain why. We will systematically classify these unannotated spectra to determine what proportion arise from redundant or non-analyte signals (isotopes, adducts, in-source fragments, multimers, contaminants, and noise) versus genuinely novel metabolites. By filtering out the former, we will assemble a clean, repository-scale molecular network: a molecular atlas that prioritizes truly unknown chemistry for downstream discovery. Finally, we will rank the unannotated spectra into a prioritized "hit list" of those whose annotation would cover the most chemical space or repository space via analogue propagation. This highlights where annotation effort pays off most.

Project Plan

Before the hackathon

1. Process GNPS with Everything Bagel + Falcon → deduplicated, repository-scale consensus spectra.

During the hackathon

1. Library search: annotate consensus spectra against public libraries.
2. Spectral library analysis: reconstruct annotation coverage year over year against the GNPS spectral library, showing the impact of growing public libraries on rediscovery of already-detected molecules.
3. DB retrieval: annotate part of the no-match spectra via rule-based and machine learning-based database retrieval tools.
4. Unannotated spectrum analysis: classify what remains: isotopes, adducts, in-source fragments, multimers, contaminants, noise, or novel metabolites.
5. Molecular Networking: build a clean, repository-scale molecular network (the molecular atlas).
6. Create hit list: List the spectra whose annotation would unlock the most chemical space or repository space via analogue propagation, ranking them so the community knows which unknowns are worth chasing first.

Technical Details

• Programming language: mostly python
• Software: GNPS (web), spectrum annotation tools (MIST-CF, SIRIUS, BUDDY, …) DB retrieval tools(SIRIUS, MIST, JESTR, ICEBERG, …)
• Datasets: All of GNPS will have been processed by EverythingBagel and Falcon. Structural databases (PubChem, Wikidata/LOTUS…), public GNPS spectral libraries

Contact information

Samer Makni
University of Antwerp, Belgium
samer.makni@uantwerpen.be

Janne Heirman
University of Antwerp, Belgium
janne.heirman@uantwerpen.be

Michael Strobel
UC Riverside, California, USA
mstro016@ucr.edu

Yasin El Abiead
BOKU University, Austria
yasin.el-abiead@boku.ac.at

Alejandro Mendoza Cantu
University of Antwerp, Belgium
alejandro.mendozacantu@uantwerpen.be