How much of natural-product chemical space has a reference spectrum? An open audit of GNPS, MassBank and MoNA against COCONUT and LOTUS
4 affiliations
- 1Independent Researcher
- 2Department of Pharmaceutical and Medicinal Chemistry
- 3Veritas University Abuja
- 4University of Abuja
Abstract
Annotation of untargeted mass spectrometry data rests on reference MS/MS spectra, yet how much of documented natural-product (NP) chemistry actually has one has been asserted rather than measured. We present SPECGAP, an open, dependency-free engine that joins NP structure databases to spectral libraries on InChIKey identity and reports two quantities with explicit denominators: structural coverage, the share of reference structures with at least one experimental MS2 spectrum, and name-recoverability, the share of spectra whose declared compound name reconciles with the reference database. Applied to the complete open experimental layer—2,745,226 MS2 spectra from MassBank, MoNA and GNPS (280,024 distinct compounds)—against 738,823 COCONUT and 227,298 LOTUS structures, SPECGAP finds that 90.34% of COCONUT and 88.16% of LOTUS have no spectrum even when stereochemistry is ignored (exact-key coverage 2.86% and 5.08%). Coverage grows sublinearly: a 19.7-fold increase in spectra bought a 2.63-fold increase in covered structures, an elasticity of 0.32, because only 7.5% of the compounds in the layer are COCONUT natural products. Composition outweighs size: GNPS added 7.4-fold fewer new compounds than MoNA but a 5.9-fold higher NP hit rate (18.8% versus 3.2%). Both measures depend on the reference database: LOTUS is 3.25-fold smaller than COCONUT yet 1.78-fold better covered, while the same spectra recover 69.54% of their names against COCONUT and 36.17% against LOTUS. On the one run with archived match-level detail, 12.1% of recoveries came from fuzzy name matching; exact matching alone gives 31.82%, and we show the fuzzy rule can accept congener names. Library joinability differs 6.5-fold, and the GNPS MGF export cannot be joined at all. All code, audit outputs and derivations are openly released.
Related
- Multimodal Representation Learning for Exploring Natural Product Chemical Space
- Lost in Chemical Space: Local Open-Source Large Language Models Display Model-Dependent Adherence to Reference Data in Molecular Generation
- Natural Product Chemistry in Sickle Cell Disease Therapeutics: Phytochemical Mechanisms, Pharmacokinetics, and Computational Docking Insights for Next-Generation Anti-Sickling Agents: A Comprehensive Review
- LANaPDB: A web-based platform for the Latin American Natural Product Database
- Strategic Applications of Single-Atom Skeletal Editing in Natural Product Synthesis
BibTeX
@misc{erumiseli2026much,
title = {{How much of natural-product chemical space has a reference spectrum? An open audit of GNPS, MassBank and MoNA against COCONUT and LOTUS}},
author = {Oshoma Erumiseli and Bila Hassan Ali and Dauda Garba},
howpublished = {ChemRxiv},
year = {2026},
month = oct,
doi = {10.26434/chemrxiv.15010472/v1},
url = {https://chemrxiv.org/doi/full/10.26434/chemrxiv.15010472/v1},
note = {CC BY},
}
RIS
TY - UNPB
TI - How much of natural-product chemical space has a reference spectrum? An open audit of GNPS, MassBank and MoNA against COCONUT and LOTUS
AU - Oshoma Erumiseli
AU - Bila Hassan Ali
AU - Dauda Garba
PY - 2026
DA - 2026/10/14
PB - ChemRxiv
DO - 10.26434/chemrxiv.15010472/v1
UR - https://chemrxiv.org/doi/full/10.26434/chemrxiv.15010472/v1
AB - Annotation of untargeted mass spectrometry data rests on reference MS/MS spectra, yet how much of documented natural-product (NP) chemistry actually has one has been asserted rather than measured. We present SPECGAP, an open, dependency-free engine that joins NP structure databases to spectral libraries on InChIKey identity and reports two quantities with explicit denominators: structural coverage, the share of reference structures with at least one experimental MS2 spectrum, and name-recoverability, the share of spectra whose declared compound name reconciles with the reference database. Applied to the complete open experimental layer—2,745,226 MS2 spectra from MassBank, MoNA and GNPS (280,024 distinct compounds)—against 738,823 COCONUT and 227,298 LOTUS structures, SPECGAP finds that 90.34% of COCONUT and 88.16% of LOTUS have no spectrum even when stereochemistry is ignored (exact-key coverage 2.86% and 5.08%). Coverage grows sublinearly: a 19.7-fold increase in spectra bought a 2.63-fold increase in covered structures, an elasticity of 0.32, because only 7.5% of the compounds in the layer are COCONUT natural products. Composition outweighs size: GNPS added 7.4-fold fewer new compounds than MoNA but a 5.9-fold higher NP hit rate (18.8% versus 3.2%). Both measures depend on the reference database: LOTUS is 3.25-fold smaller than COCONUT yet 1.78-fold better covered, while the same spectra recover 69.54% of their names against COCONUT and 36.17% against LOTUS. On the one run with archived match-level detail, 12.1% of recoveries came from fuzzy name matching; exact matching alone gives 31.82%, and we show the fuzzy rule can accept congener names. Library joinability differs 6.5-fold, and the GNPS MGF export cannot be joined at all. All code, audit outputs and derivations are openly released.
ER -