Predicting peptides bound to I‐A^g7 class II histocompatibility molecules using a novel expectation‐maximization alignment algorithm

Abstract

The useful structural features of class II MHC molecules are rarely integrated into T-cell epitope predictions. We propose an approach that applies a novel expectation-maximization algorithm to align the naturally processed peptides selected by the class II MHC I-A(g7) molecule - focusing on the five MHC-specific anchor positions. Based on the alignment profile, log of odds (LOD) scores supplemented with the Laplace plus-one pseudocounts method are applied to identify the potential T-cell epitopes. In addition, an innovative computational concept of hindering residues using statistical and structural information is developed to refine the prediction. Performance analysis by receiver operating characteristics statistics and the experimental validation of the LOD scores demonstrate the accuracy of our predictive model. Furthermore, our model successfully predicts T-cell epitopes of hen egg-white lysozyme protein antigen. Our study provides a framework for predicting T-cell epitopes in class II MHC molecules.

The peptide-binding motif of HLA-DR8 shares important structural features with other type 1 diabetes-associated alleles

“…30,31 The use of naturally processed peptides instead of synthetic peptides or phage display libraries for the identification of peptidebinding motifs to MHC II molecules have the main advantage that the peptides had been selected by the antigen-presenting cells during the antigen processing, where a number of cellular events influence peptide selection. 32 We sequenced 486 unique naturally processed DR8-bound peptides by MDLC-mESI-ITMS/MS. As expected, most proteins generating the ligands were from membrane compartments, including plasma membrane, golgi apparatus, endosomes, lysosomes, endoplasmic reticulum and secretion vesicles.…”

Do the peptide-binding properties of diabetogenic class II molecules explain autoreactivity?

“…Whether such peptides form a significant fraction of the repertoire of islet β cell epitopes that are targeted by diabetogenic T cells needs to be examined in finer detail. I-Ag7 binding motif as determined by the alignment results from the expectation-maximization alignment algorithm of Chang et al [27] and shown by WebLogo. Preferred amino acids from positions P1 to P9 are indicated by their one letter code.…”

“…Recently, we utilized the vast datasets of natural peptides to develop a computational program that efficiently predicts epitopes selected by diabetogenic class II MHC molecules [27]. This approach had several unique aspects: (i) only the naturally selected peptides isolated from APCs were used in the development of this algorithm; (ii) the physical and structural characteristics of the binding pockets were used to refine the preferred amino acid anchors; and (iii) hindering residues, i.e.…”

“…All these factors, when applied in concert, enabled this algorithm to identify the preferred binding motif for I-A g7 , as shown in Figure 1. When tested against many natural peptides, the algorithm successfully assigned high scores to high affinity binders and lower scores to peptides that bound poorly [27]. Moreover, the algorithm successfully predicted I-A g7 peptides from hen egg lysozyme that were recognized by T cells in NOD mice [27].…”

“…When tested against many natural peptides, the algorithm successfully assigned high scores to high affinity binders and lower scores to peptides that bound poorly [27]. Moreover, the algorithm successfully predicted I-A g7 peptides from hen egg lysozyme that were recognized by T cells in NOD mice [27]. Hence, in this case, predictive algorithms that have evolved from analysis of naturally processed peptides from class II MHC molecules can be powerful tools for future analysis of unknown epitopes from target islet beta cell antigens.…”