Neural network-based peak detection for untargeted metabolomics LC-MS data.

This package provides a high-level Python API for the peakonly neural network, making it easy to integrate into data analysis pipelines and Jupyter notebooks.

• Automatic peak detection using pre-trained neural networks
• Batch processing of multiple mzML files
• Simple API for integration into existing workflows
• Automatic model download on first use
• Returns results as pandas DataFrames

pip install git+https://github.com/sorenwacker/ms-peakonly

• PyTorch >= 1.2.0
• pandas
• numpy
• pymzML
• scipy
• matplotlib

from ms_peakonly import PeakOnly
from glob import glob

# Get list of mzML files
files = glob("data/*.mzML")

# Initialize (downloads models automatically on first use)
po = PeakOnly(model_dir="models/")

# Process files and get peak table
peaks = po.process(files)

# Results are returned as a pandas DataFrame
print(peaks.head())

The neural network approach identifies chromatographic peaks directly from raw LC-MS data without requiring traditional signal processing steps like smoothing or baseline correction. The model was trained on manually annotated peaks from diverse metabolomics datasets.

• Supported format: mzML files
• Files should contain centroided MS1 data

The process() method returns a pandas DataFrame containing detected peaks with columns for:

• Retention time
• m/z value
• Intensity
• Peak boundaries
• Quality scores

If you use this package in your research, please cite the original peakonly paper:

Kumler, W., et al. (2023). Neural network-based peak detection for untargeted metabolomics.

This package wraps the neural network implementation from Arseha/peakonly.

MIT License - see LICENSE for details.