MassLeague: A MS-based Compound Structure Annotation Framework Compatible with Federated Learning

A collaborative and privacy-preserving framework for compound annotation from mass spectrometry data using deep learning and federated computing.

MassLeague is a structural annotation framework for mass spectrometry (MS) that enables accurate, reproducible, and privacy-respecting compound identification across institutions. It integrates cutting-edge spectrum-centric and structure-centric deep learning engines with a decentralized, federated learning architecture.

MassLeague addresses challenges such as:

• Limited spectral libraries
• Data silos across labs
• Low reproducibility of models

• Multi Annotation Engines

  • FederEI v2: EI-MS engine with in-silico spectral generation + fingerprint prediction
  • DeepMASS v2: MS/MS engine using Spec2Vec + structure localization

• Federated Learning

  • Secure multi-party computation (SMPC)
  • HNSW-based private spectral searching
  • Distributed model optimization without data sharing

• Metabolomics: Annotation of unknown metabolites in large-scale untargeted LC/GC-MS datasets

• Toxicology: Identification of unknown compounds in complex mixtures

• Drug Discovery: Structure elucidation from MS/MS data of synthetic or natural products

• Environmental Chemistry: Compound identification in environmental samples across labs

MassLeague/
├── DeepMASS v2/                   # DeepMASS v2 MS/MS annotation engine
│   ├── (sources)                  # Configuration files for training/searching
│   ├── DeepMASS2.py               # Standalone Running DeepMASS v2
│   └── DistributedDeepMASS2.py    # Distributed Running DeepMASS v2
│   └── ReadMe.md 
│
├── FederEI v2/                    # FederEI v2 EI-MS annotation engine
│   ├── (sources)                  # Configuration files for training/searching
│   ├── FederEI2.py (.exe/.jar)    # Executable file of FederEI v2
│   └── ReadMe.md
│
├── FederalTraining/
│   ├── (sources)                  # Configuration files for training/searching
│   ├── (scripts)                  # Running scripts for training/searching
│   └── ReadMe.md
│
└── ReadMe.md

Installation instructions and example usage can be found in the respective federei and deepmass folders.

Please refer to:

• federei/README.md
• deepmass/README.md

Sihao Chang ([email protected])
Ziyao Xiong ([email protected])
Hongchao Ji ([email protected])