BoltzMatch

BoltzMatch applies restricted Boltzmann machines to improve peptide-spectrum-matching (PSM) scoring in tandem mass spectrometry for shotgun proteomics.

Key Features:

• Stochastic Neural Networks: Employs restricted Boltzmann machines (RBMs) to learn and enhance score functions that discriminate correct from incorrect peptide-spectrum matches.
• Chemically Explainable Patterns: Identifies and learns chemically explainable patterns among peak pairs in MS2 spectra, capturing the semantic context of peaks.
• Peak Augmentation and Reconstruction: Augments existing peaks based on contextual relevance and reconstructs missing expected ion peaks during internal scoring.

Scientific Applications:

• Enhanced Annotations: Achieves 50% more annotations on high-resolution MS2 data and 33% more on low-resolution MS2 data compared to XCorr at 0.1% FDR.
• Error Reduction: Maintains the same number of spectrum annotations as XCorr while reducing errors by 4- to 6-fold.
• Comparison with Other Tools: Provides 14% more annotations than Prosit (with Percolator) alone and yields a 32% increase in annotations when combined with Percolator at the same FDR.

Methodology:

Uses restricted Boltzmann machines trained on spectral data to learn peak-pair patterns and applies peak augmentation and reconstruction within its internal scoring to improve PSM discrimination.