MHCflurry

MHCflurry predicts binding affinities of peptides to major histocompatibility complex class I (MHC I) proteins to inform immunology research, vaccine design, and epitope selection.

Key Features:

• Allele-Specific Neural Networks: Employs allele-specific neural networks trained on affinity measurements to provide allele-tailored binding predictions.
• Novel Peptide Encoding Scheme: Uses a peptide encoding scheme that improves prediction accuracy for peptides of varying lengths, including non-9-mer peptides.
• Performance and Speed: Benchmarking on mass spectrometry–identified ligands showed superior performance versus NetMHC 4.0 and NetMHCpan 3.0—especially on non-9-mer peptides—and reported prediction throughput of over 7,000 predictions per second, approximately 396× faster than NetMHCpan 4.0.
• Model Selection and Accuracy: Version 1.2.0 uses mass spectrometry datasets for model selection and reports accuracy competitive with leading tools including NetMHCpan 4.0.

Scientific Applications:

• Vaccine design: Prioritizes MHC I peptide candidates for vaccine antigen design by predicting peptide–MHC binding affinities.
• Immunotherapy: Supports identification of peptide targets for therapeutic interventions by ranking binding affinities.
• Epitope identification: Aids selection of candidate epitopes from peptide libraries or mass spectrometry ligand lists through affinity prediction.

Methodology:

Trains allele-specific neural networks on affinity measurement datasets, applies a peptide encoding scheme to handle non-9-mer peptides, and incorporates mass spectrometry datasets for model selection.