NetMHC

NetMHC predicts peptide binding affinities to HLA alleles using artificial neural networks to identify peptides likely to bind MHC molecules.

Key Features:

• Neural network encoding: Uses artificial neural networks integrating sparse encoding, Blosum encoding, and hidden Markov model (HMM)-derived inputs to capture higher-order sequence correlations.
• Mutual information analysis: Employs mutual information calculations to identify higher-order sequence correlations in peptides that bind strongly to HLA molecules.
• Peptide length versatility: Trained on 9-mer peptide data and adapted to predict binding for 8-, 10-, and 11-mer peptides without direct training on those lengths.
• Gibbs motif sampler for MHC class II: Applies a Gibbs motif sampler to locate weak sequence motifs and characterize them as weight matrices for MHC class II binding predictions.

Scientific Applications:

• T-cell epitope prediction: Predicts peptides that bind specific HLA alleles to aid identification of potential T-cell epitopes, including epitopes relevant to hepatitis C virus.
• Vaccine design and immunotherapy: Guides selection of immunogenic peptides for vaccine development and supports design of personalized cancer immunotherapies.

Methodology:

Artificial neural networks trained on peptide binding data (originally 9-mer), multiple sequence-encoding schemes (sparse, Blosum, HMM-derived inputs), mutual information calculations to detect higher-order correlations, and Gibbs sampling/motif extraction for MHC class II producing weight matrices.