An automated prediction of MHC class I-binding peptides based on positional scanning with peptide libraries

Abstract

Specificities of three mouse major histocompatibility complex (MHC) class I molecules, Kb, Db, and Ld, were analyzed by positional scanning using combinatorial peptide libraries. The result of the analysis was used to create a scoring program to predict MHC-binding peptides in proteins. The capacity of the scoring was then challenged with a number of peptides by comparing the prediction with the experimental binding. The score and the experimental binding exhibited a linear correlation but with substantial deviations of data points. Statistically, for approximately 80% of randomly chosen peptides, MHC-binding capacity could be predicted within one log concentration of peptides for a half-maximal binding. Known cytotoxic T-lymphocyte epitope peptides could be predicted, with a few exceptions. In addition, frequent findings of MHC-binding peptides with incomplete or no anchor amino acid(s) suggested a substantial bias introduced by natural antigen processing in peptide selection by MHC class I molecules.

Enhancement to the RANKPEP resource for the prediction of peptide binding to MHC molecules using profiles

“…Prediction of peptide-MHC binding using profiles: comparison with other methods Prediction of peptide-MHC binding is important for the anticipation of T-cell epitopes and so determination of peptides that can bind to MHC molecules has been approached by a large array of methods including sequence patterns (Sette et al 1989), motif-matrices (De Groot et al 1997Rammensee et al 1999), quantitative matrices (QM) (Guan et al 2003;Hammer et al 1994;Parker et al 1994;Stryhn et al 1996;Udaka et al 2000), virtual quantitative matrices (VQM) (Raddrizzani and Hammer 2000;Sturniolo et al 1999), artificial neural networks (ANN) (Adams and Koziol 1995;Brusic et al 1998a;Gulukota et al 1997;Honeyman et al 1998); hidden Markov motifs (HMM) (Mamitsuka 1998;Udaka et al 2002); structural peptide threading (SPT) (Altuvia et al 1997;Schueler-Furman et al 2000;Swain et al 2001), support vector machine (SVM) algorithms (Donnes and Elofsson 2002;Zhao et al 2003) and stepwise discriminant analysis meta-algorithm (SDA) (Mallios 1999). QM and VQM methods are derived from actual binding experiments, whereas SPT is an entirely computer-based method that relies on the evaluation of peptide fit into the binding groove, and despite its great potential is currently still under development.…”

Prediction of MHC class I binding peptides using profile motifs

“…Prediction matrices of Parker et al [26] were generated from a limited set of peptides, perhaps explaining a recent report [33] describing poor correspondence between the predicted MHC binding peptides and those determined experimentally. Quantitative matrices have also been derived from positional scanning combinatorial peptide libraries (PSCPL) [27,28], where all possible peptides of a given length are represented by sets of sublibraries and in each sublibrary, one amino acid is kept fixed whereas the remaining positions contain mixtures of all amino acids. Unfortunately, to date, prediction of pMHCI binding using these PSCPL-derived matrices is not freely accessible.…”

Identifying MHC Class I Epitopes by Predicting the TAP Transport Efficiency of Epitope Precursors

“…This systematic increase can be explained by the fact that, due to similar peptide preferences of TAP and MHC, the likelihood for a peptide with high affinity to HLA-A0201 to possess at least a decent TAP transport rate is higher than for a peptide with low affinity to HLA-A0201. We repeated the same two-step prediction for several mouse MHC-I alleles using scoring matrices for the MHC-I affinity prediction that were measured by Udaka et al (27). Unfortunately, the number of epitopes available per allele is small (ranging from 9 to 21).…”