NetMHCpan

NetMHCpan is a computational tool and web server for predicting the binding of peptides to MHC class I molecules across a broad range of species. It belongs to the NetMHC family and implements a pan-specific approach that can make predictions for alleles without dedicated binding data by exploiting similarities in the MHC peptide-binding groove.

The method uses artificial neural networks trained on large datasets of peptide-MHC binding affinities and naturally presented (eluted) ligands. Alleles are represented by a pseudo-sequence consisting of key residues in the peptide-binding groove, enabling the model to generalize to uncharacterized variants.

Predictions typically include a numeric binding affinity (IC50 in nM) and a percentile rank relative to a set of random natural peptides. NetMHCpan supports many human and non-human MHC class I molecules (HLA-A, -B, -C, and orthologs such as murine H-2). It is widely used for epitope discovery, vaccine design, and cancer immunotherapy research.

Input to the tool consists of peptide sequences (commonly 8–11 residues) and a specified MHC allele; output provides predicted binding strength, binder classifications, and confidence measures. The model is updated in successive versions to improve accuracy by incorporating eluted-ligand data and expanding allele coverage.

Limitations include that predictions do not guarantee immunogenicity and depend on the representation of the allele in training data. Experimental validation is recommended to confirm predicted epitopes.