DeepRescore

DeepRescore enhances peptide-spectrum match (PSM) rescoring in immunopeptidomics by integrating deep learning-derived retention time and MS/MS spectral features to improve identification of MHC-binding peptides and neoantigens from mass spectrometry (MS) data.

Key Features:

• Deep learning integration: Utilizes deep learning predictions of retention time and MS/MS spectra to derive features that are combined with traditional features for rescoring PSMs.
• Support for multiple search engines: Accepts initial identification results from MS-GF+, Comet, X!Tandem, and MaxQuant.
• Input formats: Processes MS/MS data in Mascot Generic Format (MGF) alongside search-engine identification outputs.
• Improved sensitivity and reliability: Incorporation of deep learning-derived features increases sensitivity and reliability of MHC-binding peptide and neoantigen identification compared to existing methods.

Scientific Applications:

• Immunopeptidomics: Enhances identification of MHC-binding peptides from MS-based immunopeptidomics datasets.
• Neoantigen discovery: Improves detection of candidate neoantigens for cancer immunotherapy studies.
• Antigen presentation and immune-response analysis: Supports studies of antigen presentation relevant to vaccine development and immune-response characterization.

Methodology:

Inputs comprise MS/MS data in Mascot Generic Format (MGF) and initial identification results from supported search engines; a rescoring algorithm combines traditional PSM features with deep learning-derived retention time and MS/MS spectral predictions, and performance was validated on two public immunopeptidomics datasets.