PB-Net

PB-Net automates chromatographic peak boundary detection and area integration in Multiple Reaction Monitoring (MRM) mass spectrometry for quantitative proteomics and glycoproteomics.

Key Features:

• Automatic Peak Integration: PB-Net leverages sequential deep learning to perform fully automatic chromatographic peak integration, reducing the need for manual expert annotations.
• Sequential Neural Network Architecture: Built on sequential neural network architectures, PB-Net detects peak boundaries and integrates peak areas from chromatographic traces with high precision.
• Training on Extensive Dataset: The model was trained on over 170,000 expert-annotated peaks from MS transitions, including peptides and intact glycopeptides across a wide dynamic range.

Scientific Applications:

• Mass Spectrometry Data Analysis: Provides accurate and reliable peak integration for MS-based proteomics and glycoproteomics data analysis.
• High-Throughput MS Experiments: Enables rapid and reproducible quantification in high-throughput MRM experiments.

Methodology:

Test predictions are stored in test_preds.pkl and test_preds_ref.pkl as peak start/end probability outputs and can be generated with run.py using pre-trained models; test_input.pkl contains a list of peak instances where each entry is a tuple of features (X), labels (y), metainfo, and identifiers (sample name and peak name).