SMSNet

SMSNet predicts de novo peptide sequences from tandem mass spectrometry spectra using a deep learning model to enable identification of novel peptides such as HLA antigens and phosphopeptides.

Key Features:

• Deep Learning Framework: Employs a deep learning model for de novo peptide sequencing and reports over 95% amino acid accuracy while maintaining high identification coverage.
• Sequence-Mask-Search Hybrid Approach: Integrates a Sequence-Mask-Search framework that combines sequence prediction with search capabilities to discover peptides beyond existing database constraints.
• Comprehensive Training Datasets: Trained on WCU-MS-BEST, DeepNovo, ProteomeTools, and WCU-MS-ALL datasets to support robust model performance across diverse peptide sequences.
• Dual Model Availability: Provides two models within the SMSNet framework to address different analysis needs.

Scientific Applications:

• Uncharacterized Peptide Identification: Enabled identification of over 10,000 previously uncharacterized HLA antigens and phosphopeptides.
• Integration with Database Searches: When integrated with database-search methods, expands peptide identification coverage by nearly 30%.
• Tumor Neoantigen Discovery: Facilitates discovery of tumor-specific neoantigens relevant to cancer immunotherapy research.
• Antibody Sequencing: Supports sequencing of antibodies for therapeutic antibody development and characterization.
• Proteome Characterization: Assists characterization of proteomes from non-model organisms.

Methodology:

Uses a deep learning model implementing a Sequence-Mask-Search approach to predict amino acid sequences from mass spectrometry data; trained on WCU-MS-BEST, DeepNovo, ProteomeTools, and WCU-MS-ALL; applied to datasets including MassIVE accession MSV000080527 and PRIDE accession PXD009227, with predictions made available via FigShare.