lbnl-metabolomics/BLAMSS_v1.0

Berkeley Lab Atlas of Mass Spectra Standards 1.0 (BLAMSS 1.0)

Bridging the gap between chemical analysis and computational modeling in metabolomics.

---

Dataset Overview

The rapid proliferation of Artificial Intelligence (AI) and Machine Learning (ML) in metabolomics is currently bottlenecked by a scarcity of high-quality, experimentally validated training datasets. BLAMSS 1.0 addresses this by providing open access to large-scale data derived from authentic chemical standards.

Unlike crowd-sourced repositories, this dataset represents a centralized, homogeneous collection of high-confidence (Metabolomics Standards Initiative Level 1) identifications. It is designed specifically to serve as a "ground truth" reference for training generative models, spectral predictors, and retention time forecasters.

Methodology & Acquisition

This dataset was generated by the Berkeley Lab through the systematic analysis of over 5,000 authentic metabolite standards. The data acquisition pipeline was strictly standardized to ensure high transferability:

• Instrumentation: Thermo Orbitrap Mass Spectrometer coupled with an Agilent 1290 UHPLC system.
• Chromatography: Samples were run on two complementary standardized columns to maximize coverage of chemical space:
  • Reverse Phase (C18): For lipids and non-polar metabolites.
  • HILIC (Hydrophilic Interaction Liquid Chromatography): For polar and ionic metabolites.
• Mass Spectrometry Settings:
  • Polarity: Both Positive (+) and Negative (-) ionization modes.
  • Collision Energies (CE): To support robust spectral prediction, standards were fragmented at multiple collision energies, providing a fragmentation landscape for each molecule.

Dataset Structure

To maximize utility for the computational community, the dataset is provided in two formats:

1. AI-Ready Parquet (ema_standards_everything_with_ms2.parquet)

A highly optimized, columnar file ideal for immediate loading into Python/Pandas/PyTorch pipelines.

• Rows: Individual spectral snapshots.
• Schema:
  • ms2_mz / ms2_intensity: Arrays containing the fragmentation spectrum.
  • mz: Precursor m/z.
  • rt_peak: Experimental retention time.
  • adduct: Precursor adduct type (e.g., [M+H]+, [M-H]-).
  • inchikey / inchi: Chemical identifiers for exact structure mapping.
  • ms_settings: Specific collision energy used for the scan.
  • file_name: The rawdata filename that this molecule was observed in.

2. Raw Data (/rawdata/*.mzML)

To link the raw data to the AI ready parquet file, the original open-standard .mzML files are included for researchers developing raw signal processing algorithms, peak picking software, or those requiring the full chromatographic context.

Impact & Applications

BLAMSS 1.0 democratizes access to high-confidence metabolomics data, facilitating:

1. Next-Generation AI Model Development:
   • De novo spectral prediction (Graph Neural Networks, Transformers).
   • Retention Time (RT) forecasting and transfer learning.
   • Substructure inference from MS/MS spectra.
2. Benchmarking: A gold-standard test set for evaluating existing metabolite annotation and peak picking software.
3. Cross-Laboratory Transferability: Includes robust protocols for RT correction using internal standards, allowing these retention times to be mapped to external datasets collected on similar hardware.
4. Raw data mining: Includes the raw data so studies to understand in source fragmentation and adduct formation can be implemented.

Citation

If you use BLAMSS 1.0 in your research, please cite this dataset