Neural network-based prediction of candidate T-cell epitopes

Nat Biotechnol. 1998 Oct;16(10):966-9. doi: 10.1038/nbt1098-966.


Activation of T cells requires recognition by T-cell receptors of specific peptides bound to major histocompatibility complex (MHC) molecules on the surface of either antigen-presenting or target cells. These peptides, T-cell epitopes, have potential therapeutic applications, such as for use as vaccines. Their identification, however, usually requires that multiple overlapping synthetic peptides encompassing a protein antigen be assayed, which in humans, is limited by volume of donor blood. T-cell epitopes are a subset of peptides that bind to MHC molecules. We use an artificial neural network (ANN) model trained to predict peptides that bind to the MHC class II molecule HLA-DR4(*0401). Binding prediction facilitates identification of T-cell epitopes in tyrosine phosphatase IA-2, an autoantigen in DR4-associated type1 diabetes. Synthetic peptides encompassing IA-2 were tested experimentally for DR4 binding and T-cell proliferation in humans at risk for diabetes. ANN-based binding prediction was sensitive and specific, and reduced the number of peptides required for T-cell assay by more than half, with only a minor loss of epitopes. This strategy could expedite identification of candidate T-cell epitopes in diverse diseases.

Publication types

  • Research Support, Non-U.S. Gov't

MeSH terms

  • Amino Acid Sequence
  • Epitopes / chemistry*
  • Histocompatibility Antigens Class I / immunology
  • Homozygote
  • Humans
  • Molecular Sequence Data
  • Neural Networks, Computer*
  • Sequence Homology, Amino Acid
  • T-Lymphocytes / immunology*


  • Epitopes
  • Histocompatibility Antigens Class I