deepAntigen

deepAntigen integrates 3D genome information and machine learning to prioritize MHC-I neoantigens for personalized cancer immunotherapy.

Key Features:

• 3D genome integration: Incorporates spatial genomic (3D) information of DNA loci to inform neoantigen immunogenicity predictions.
• Ensemble pMHC-I coding strategy: Employs an ensemble coding scheme for peptide–MHC class I (pMHC-I) representations.
• Group feature selection-based deep sparse neural network (DNN-GFS): Uses a deep sparse neural network with group feature selection to prioritize neoantigen candidates.
• Additional machine learning methods: Implements multiple other machine learning approaches alongside DNN-GFS for comparative analysis.
• Spatial distribution insight: Leverages reported differences in genomic spatial distribution between immunopositive and immunonegative MHC-I neoantigens.
• Improved prioritization versus sequence-only methods: Optimizes neoantigen ranking by combining 3D genomic features with advanced coding and modeling beyond sequence-based methods.

Scientific Applications:

• Neoantigen prioritization: Prioritizes candidate MHC-I neoantigens derived from somatic cancer mutations for downstream experimental validation.
• Immunogenicity prediction: Aids prediction of T cell–recognizable immunogenic peptides presented by MHC-I.
• Personalized cancer immunotherapy design: Supports selection of neoantigen targets for personalized cancer immunotherapies and vaccine development.
• Spatial genomics analysis of immunogenicity: Enables analysis of spatial genomic patterns associated with immunopositive versus immunonegative neoantigens.

Methodology:

Integrates 3D genome information with an ensemble pMHC-I coding strategy and applies a group feature selection–based deep sparse neural network (DNN-GFS), alongside other machine learning methods.