Alpha-Frag

Alpha-Frag predicts fragment ion presence in MS/MS spectra using a deep neural network to improve peptide identification accuracy in mass spectrometry-based proteomics.

Key Features:

• Deep Learning-Based Prediction: Uses a deep neural network to predict the presence of fragment ions from peptides, modeling the task as a multi-label classification problem.
• Training and Performance: Trained on the ProteomeTools dataset with an average intersection over union (IoU) score greater than 0.7 and reported to outperform existing benchmarks across validation datasets.
• Fragment Presence Similarity Calculation: Calculates fragment presence similarity by comparing predicted fragment spectra with experimental MS/MS data.
• Integration into Statistical Validation: Incorporates predicted fragment presence as an additional score in peptide statistical validation, improving identification rates up to 26.8% at 0.1% FDR for DDA and 21.6% at 1% FDR for DIA.
• Use of Spectral Features: Considers both mass-to-charge ratio (m/z) and intensity data when modeling peptide fragmentation patterns.

Scientific Applications:

• Data-Dependent Acquisition (DDA) workflows: Improves peptide identification reliability in DDA workflows by providing more precise fragment presence predictions.
• Data-Independent Acquisition (DIA) workflows: Enhances peptide identification and validation in DIA workflows through fragment presence similarity scoring.
• Peptide statistical validation: Provides an additional predictive score to augment peptide statistical validation and increase identification rates.

Methodology:

Models fragment presence prediction as a multi-label classification problem using a deep neural network trained on the ProteomeTools dataset, leveraging m/z and intensity data and computing fragment presence similarity between predicted and experimental spectra.