usiGrabber

usiGrabber is a tool for building large-scale mass spectrometry-based proteomics datasets from publicly available data. It parses peptide spectrum matches (PSMs) and associated metadata from mzIdentML files from PRIDE, stores them in a queryable database, and provides tools to download and export the corresponding raw spectra for downstream machine learning use.

Repository Overview

This repository contains two packages that together implement the full dataset construction pipeline:

Package	Description	Documentation
usiGrabber	Extracts PSMs and metadata from PRIDE into a PostgreSQL database	src/usigrabber/README.md
Spectrum Toolkit	Queries the database and downloads raw spectra for use in machine learning tasks	src/spectrum_toolkit/README.md

Proof-of-Concept: Curating a dataset for retraining a binary phosphorylation classifier

As a proof of concept, we used usiGrabber to curate a dataset for retraining a binary phosphorylation classifier based on the AHLF model architecture. Adjustments to the original code and the weights of our retrained model are available in the AHLF-fork repository.

General Setup

1. Install prerequisites

   • uv
   • Docker (optional, recommended for PostgreSQL)

2. Install dependencies and set up pre-commit hooks

   uv sync
   uv run pre-commit install

3. Configure environment variables

   • Copy .env.sample to .env and adjust as needed (database URL, credentials).

For detailed database setup and build commands, see src/usigrabber/README.md.

Data Availability

The data is available on Zenodo:

• Main record containing the training dataset and model weights for trained AHLF model: https://zenodo.org/records/18853258
• The usiGrabber database tables are split across multiple records:
  • peptide_spectrum_matches table: https://zenodo.org/records/18890370
  • psm_peptide_evidence table: https://zenodo.org/records/18864164
  • Other, smaller tables: https://zenodo.org/records/18873214

Refer to our database import/export guide on how to reassemble the database from these records.

Citation

usiGrabber: Automating the curation of proteomics spectra data at scale, making large datasets ready for use in machine learning systems
Georg Auge, Matthis Clausen, Konstantin Ketterer, Jacob Schaefer, Nils Schmitt, Tom Altenburg, Yannick Hartmaring, Hendrik Raetz, Christoph N. Schlaffner, Bernhard Y. Renard
doi: https://doi.org/10.64898/2026.03.15.711873