DeepRescore: Leveraging Deep Learning to Improve Peptide Identification in Immunopeptidomics

Area of Science:

• Immunology
• Proteomics
• Bioinformatics

Background:

• Mass spectrometry-based immunopeptidomics relies on database search engines for identifying MHC-binding peptides.
• Current methods suffer from high false positive rates and low sensitivity due to an inflated search space in immunopeptidomics.
• Enzymatic digestion is not typically used in immunopeptidomics, unlike standard proteomics.

Purpose of the Study:

• To develop a post-processing tool, DeepRescore, to improve the sensitivity and reliability of peptide identification in immunopeptidomics.
• To leverage deep learning predictions to enhance peptide-spectrum matching.
• To address the limitations of existing methods in immunopeptide data analysis.

Main Methods:

• Developed DeepRescore, a post-processing tool integrating deep learning predictions (retention time, MS/MS spectra) with existing features.
• Applied DeepRescore to rescore peptide-spectrum matches in immunopeptidomics data.
• Utilized NextFlow and Docker for tool development and accessibility.

Main Results:

• Rescoring with DeepRescore significantly increased the sensitivity and reliability of MHC-binding peptide and neoantigen identifications.
• The performance improvement was largely attributed to the deep learning-derived features.
• Demonstrated enhanced identification accuracy on two public immunopeptidomics datasets.

Conclusions:

• DeepRescore offers a substantial improvement over existing methods for immunopeptide identification.
• Deep learning-based features are crucial for enhancing the accuracy of immunopeptidomics analysis.
• The developed tool provides a more sensitive and reliable approach for identifying MHC-binding peptides and neoantigens.