Alpha-Tri

Alpha-Tri improves peptide identification in data-independent acquisition (DIA) mass spectrometry by integrating predicted, measured, and correlation spectra with a neural network to enhance spectral library matching.

Key Features:

• Triple-Spectrum Approach: Integrates a Predicted Spectrum generated by Prosit encompassing all possible fragment ion intensities, a Measured Spectrum derived from the apex of chromatograms for potential fragment ions, and a Correlation Spectrum reflecting presence probabilities of fragment ions to compensate for lost precursor-to-fragment ion linkage in DIA.
• Neural Network Model: Uses a deep neural network to compute intensity similarity scores from the triple-spectra to improve peptide identification beyond traditional library matches.
• Post-Processing Score Integration: Appends the intensity similarity score to initial identification scores from DIA-NN to increase identification confidence while maintaining the same false discovery rate (FDR).

Scientific Applications:

• HeLa Dataset Evaluation: Demonstrated 3.0-7.2% improvements in peptide detections at 1% FDR across gradient lengths of 0.5 to 2 hours.
• LFQbench Dataset Analysis: Identified more peptides and proteins within valid ratio ranges by up to 8.6% and 7.6%, respectively, compared with DIA-NN alone.
• Enhanced Spectral Library Matching: Improves reliability of spectral library matching in complex biological samples.
• Improved Quantification in Mixed-Species Datasets: Enhances quantification accuracy for studies of protein expression and interaction dynamics.

Methodology:

Predicted spectra generated by Prosit and measured spectra from chromatogram apexes are combined with a correlation spectrum of fragment presence probabilities; a deep neural network computes intensity similarity scores from these triple-spectra which are appended to DIA-NN identification scores to maintain FDR.