Prediction of binding to MHC class I molecules

Abstract

The binding of antigenic peptide sequences to major histocompatibility complex (MHC) molecules is a prerequisite for stimulation of cytotoxic T cell responses. Neural networks are here used to predict the binding capacity of polypeptides to MHC class I molecules encoded by the gene HLA-A ∗ 0201. Given a large database of 552 nonamers and 486 decamers and their known binding capacities, the neural networks achieve a predictive hit rate of 0.78 for classifying peptides which might induce an immune response (good or intermediate binders) vs. those which cannot (weak or non-binders). The neural nets also depict specific motifs for different binding capacities. This approach is in principle applicable to all MHC class I and II molecules, given a suitable set of known binding capacities. The trained networks can then be used to perform a systematic search through all pathogen or tumor antigen protein sequences for potential cytotoxic T lymphocyte epitopes.