MassBank Data for Joint Annotation of Small Molecules Using LC-MS

A SQLite database built from MassBank release 2020.11 supports experiments for joint structural annotation of small molecules. The dataset includes simulated and scored tandem mass spectrometry spectra, candidate structures, and pre-computed substructure fingerprints. It was created by Eric Bach of Aalto University for the associated research publication.

Use Cases

• Training machine learning models for small molecule annotation based on mass spectrometry and retention order data.
• Benchmarking spectral scoring algorithms like MetFrag and SIRIUS using the provided candidate sets and scores.
• Reproducing the experiments from the associated research paper on joint structural annotation.
• Developing or applying the Structure Support Vector Machine (SSVM) model for metabolite identification.

Strengths

• Built from a specific, named public repository (MassBank release 2020.11).
• Includes multiple processed components: a core SQLite database, simulated spectra, pre-computed scores from two software tools, and substructure fingerprints.
• Accompanied by processing scripts to reproduce the database, supporting reproducibility.

Limitations

• Row count and overall data scale are unknown, which may limit suitability assessment.
• Column-level documentation is absent; field semantics must be inferred from the provided README files after download.
• Last update date is unknown; freshness unverified.

Provenance

Source

MassBank release 2020.11, processed via the massbank2db Python package.

Collection Method

Data was compiled and processed using scripts provided in the dataset package; includes simulated spectra from CFM-ID and scored outputs from MetFrag and SIRIUS software.