Chapter

Artificial Immune Systems

Volume 3239 of the series Lecture Notes in Computer Science pp 217-225

MHC Class I Epitope Binding Prediction Trained on Small Data Sets

  • Claus LundegaardAffiliated withCenter for Biological Sequence Analysis, BioCentrum, Technical University of Denmark
  • , Morten NielsenAffiliated withCenter for Biological Sequence Analysis, BioCentrum, Technical University of Denmark
  • , Kasper LamberthAffiliated withDepartment of Experimental Immunology, Institute of Medical Microbiology and Immunology, University of Copenhagen Denmark
  • , Peder WorningAffiliated withCenter for Biological Sequence Analysis, BioCentrum, Technical University of Denmark
  • , Christina Sylvester-HvidAffiliated withDepartment of Experimental Immunology, Institute of Medical Microbiology and Immunology, University of Copenhagen Denmark
  • , Søren BuusAffiliated withDepartment of Experimental Immunology, Institute of Medical Microbiology and Immunology, University of Copenhagen Denmark
  • , Søren BrunakAffiliated withCenter for Biological Sequence Analysis, BioCentrum, Technical University of Denmark
  • , Ole LundAffiliated withCenter for Biological Sequence Analysis, BioCentrum, Technical University of Denmark

* Final gross prices may vary according to local VAT.

Get Access

Abstract

The identification of potential T-cell epitopes is important for development of new human or vetenary vaccines, both considering single protein/subunit vaccines, and for epitope/peptide vaccines as such. The highly diverse MHC class I alleles bind very different peptides, and accurate binding prediction methods exist only for alleles were the binding pattern have been deduced from peptide motifs. Using empirical knowledge of important anchor positions within the binding peptides dramatically reduces the number of peptides needed for reliable predictions. We here present a general method for predicting peptides binding to specific MHC class I alleles. The method combines advanced automatic scoring matrix generation with empirical position specific differential anchor weighting. The method leads to predictions with a comparable or higher accuracy than other established prediction servers, even in situations where only very limited data are available for training.