IEDB — Immune Epitope Database

At a glance

Labels and annotations

Per-record assay type and outcome (binding affinity in nM, elution, T-cell response), MHC restriction, source organism and reference. The IEDB threshold convention for affinity: <50 nM high, <500 nM intermediate, <5000 nM low.

Details

Why the skew matters

• HLA-A*02:01 alone accounts for 24.38% of datasets in the IEDB automated benchmark; the seven most frequent alleles account for more than half.
• 66.24% of class I molecules have six datasets or fewer.
• A model trained on this distribution is weakest exactly for patients whose alleles were never studied.

Affinity data are a different question

• An IC50 measured on purified HLA says nothing about whether the cell makes the source protein, whether the proteasome cuts there, or whether TAP transports the peptide.

Models trained or evaluated on it

• training NetMHCpan-4.1 Health Tech, Technical University of Denmark (Nielsen lab) — binding-affinity measurements and eluted-ligand deposits
• training NetMHCIIpan-4.0 Health Tech, Technical University of Denmark (Nielsen lab)
• training MHCflurry 2.0 O'Donnell, Rubinsteyn and Laserson (Mount Sinai / OpenVax)
• training MHCnuggets Karchin lab, Johns Hopkins University

Sources

1. IEDB
2. Trevizani R et al. A comprehensive analysis of the IEDB MHC class-I automated benchmark. Brief Bioinform 2022