Bridging MS2 Spectra and Chemical Space: Advances in Spectral Similarity, Molecular Retrieval, and De Novo Structure Discovery
6 affiliations
- 1Centre for Digitalisation and Digitality
- 2Düsseldorf University of Applied Sciences
- 3Institute of Organic Chemistry and Biochemistry
- 4Czech Academy of Sciences
- 5Czech Institute of Informatics, Robotics and Cybernetics
- 6Czech Technical University
Abstract
Mass spectrometry-based metabolomics increasingly relies on computational methods to translate information-rich MS/MS fragmentation patterns into chemical insight, yet assigning chemical identities to spectra ("annotation") remains a major bottleneck because reference libraries cover only a small fraction of plausible compounds. In this review, we frame the field through the lens of two interacting representation spaces: spectral space, where MS/MS data reside, and chemical space, where molecules are represented by fingerprints, descriptors, or learned embeddings. We discuss how different design choices induce different geometries, what it means for spectra or molecules to be "close", and survey methods for representing and comparing spectra, representing molecules, and mapping between both spaces. We highlight emerging deep-learning trends for cross-modal alignment and retrieval, and we outline where the analogy to text-image learning breaks due to peak ambiguity, noise, and the one-to-many nature of spectrum-to-structure relationships.
Related
- Omics data discovery agents: Agent-supported retrieval, reanalysis, and synthesis of published omics data
- Omni-ARMD: an integrated genomic and clinical knowledgebase bridging clinical phenotypes and molecular determinants of antimicrobial resistance
- Molecular Dynamics Simulations Reveal a Broadly Open Solution Ensemble for the DNA Holliday Junction: Mg2+ Crossover Bridging and Weak Force-Field Sensitivity
- Special Issue “Molecular Mechanism Discovery of the Bioactive Phytochemicals against Different Diseases Based on Network Pharmacology and Molecular Docking, 2nd Edition”
- Discovery of a molecular glue inhibitor that stabilises a non-productive Kalirin-Rac1 complex
BibTeX
@misc{pollmann2026bridging,
title = {{Bridging MS2 Spectra and Chemical Space: Advances in Spectral Similarity, Molecular Retrieval, and De Novo Structure Discovery}},
author = {Julian Pollmann and Roman Bushuiev and Anton Bushuiev and Tomáš Pluskal and Florian Huber},
howpublished = {ChemRxiv},
year = {2026},
month = sep,
doi = {10.26434/chemrxiv.15000536/v3},
url = {https://chemrxiv.org/doi/full/10.26434/chemrxiv.15000536/v3},
note = {CC BY},
}
RIS
TY - UNPB
TI - Bridging MS2 Spectra and Chemical Space: Advances in Spectral Similarity, Molecular Retrieval, and De Novo Structure Discovery
AU - Julian Pollmann
AU - Roman Bushuiev
AU - Anton Bushuiev
AU - Tomáš Pluskal
AU - Florian Huber
PY - 2026
DA - 2026/09/09
PB - ChemRxiv
DO - 10.26434/chemrxiv.15000536/v3
UR - https://chemrxiv.org/doi/full/10.26434/chemrxiv.15000536/v3
AB - Mass spectrometry-based metabolomics increasingly relies on computational methods to translate information-rich MS/MS fragmentation patterns into chemical insight, yet assigning chemical identities to spectra ("annotation") remains a major bottleneck because reference libraries cover only a small fraction of plausible compounds. In this review, we frame the field through the lens of two interacting representation spaces: spectral space, where MS/MS data reside, and chemical space, where molecules are represented by fingerprints, descriptors, or learned embeddings. We discuss how different design choices induce different geometries, what it means for spectra or molecules to be "close", and survey methods for representing and comparing spectra, representing molecules, and mapping between both spaces. We highlight emerging deep-learning trends for cross-modal alignment and retrieval, and we outline where the analogy to text-image learning breaks due to peak ambiguity, noise, and the one-to-many nature of spectrum-to-structure relationships.
ER -