NetMHC-3.0

NetMHC-3.0 predicts MHC class I–peptide binding affinities for peptides 8–11 amino acids long across human, mouse, and monkey MHC alleles.

Key Features:

• Training data and models: Uses quantitative peptide data from the Immune Epitope Database (IEDB) and elution data from SYFPEITHI, with artificial neural networks trained on 55 MHC alleles (43 human, 12 non-human) and position-specific scoring matrices (PSSMs) for an additional 67 HLA alleles.
• Prediction outputs: Artificial neural network predictions are reported as IC50 values and PSSM-based predictions as log-odds likelihood scores.
• Performance and validation: Validated on newly published affinity data non-redundant with the training set and shown to perform well across viral proteomes including SARS, Influenza, and HIV with an average confirmation rate of 75–80% for predicted binders.

Scientific Applications:

• Vaccine development: Prioritizes MHC class I peptide candidates by predicted binding affinity to support antigen selection for vaccines.
• Disease pathogenesis studies: Identifies potential MHC-bound peptides to study immune recognition in infectious and disease contexts.
• Personalized medicine: Supports identification of individual-specific MHC-binding peptides for neoantigen selection and immunotherapy research.
• Comparative immunogenetics: Enables cross-species analyses of epitope repertoires by covering human, mouse, and monkey MHC alleles.

Methodology:

Artificial neural networks trained on quantitative peptide data from IEDB and elution data from SYFPEITHI for 55 MHC alleles, PSSMs applied to 67 additional HLA alleles, predictions reported as IC50 values (ANN) or log-odds scores (PSSM), and validation performed against non-redundant published affinity data.