DeepHLApan

DeepHLApan predicts high-confidence neoantigens by integrating human leukocyte antigen (HLA)–mutant peptide binding prediction and peptide–MHC (pMHC) immunogenicity assessment using deep learning.

Key Features:

• Dual-Model Approach: Implements separate deep learning models for binding prediction and for immunogenicity evaluation of peptide-HLA complexes.
• Binding Model: Assesses the likelihood of mutant peptides being presented by HLAs using a dedicated deep learning binding model.
• Immunogenicity Model: Evaluates the potential immune response elicited by peptide–MHC (pMHC) complexes with a complementary deep learning model.
• Improved Prediction Precision: Combines binding and immunogenicity assessments to improve neoantigen prediction precision relative to methods relying solely on HLA–peptide binding affinity.
• Benchmark Evaluation: The binding model was evaluated against Immune Epitope Database (IEDB) benchmark datasets and an independent mass spectrometry dataset.
• Clinical Relevance: Has been applied to mutations associated with pre-existing T-cell responses to identify clinically relevant neoantigens.

Scientific Applications:

• Cancer Immunotherapy: Identification of high-confidence neoantigens to inform immunotherapy research.
• Personalized Cancer Vaccines: Prioritization of neoantigen candidates for development of personalized vaccine strategies.
• Target Selection for Therapies: Selection of optimal neoantigen targets for targeted therapeutic intervention based on predicted presentation and immunogenicity.

Methodology:

Uses two deep learning models—a binding model and an immunogenicity model—to analyze binding affinity data and immunogenicity signals, with the binding model evaluated on IEDB benchmark datasets and an independent mass spectrometry dataset.