Prediction Model of MHC Class-II Binding Peptide Motifs Using Sequence Weighting Method for Vaccine Design

Identification of MHC class-II restricted epitope is an important goal in peptide based vaccine and diagnostic development. Currently, immuno informatics can circumvent conventional time-consuming and laborious experimental techniques of overlapping peptides from protein to epitope identification. However, prediction of MHC class-II epitope is difficult due to variable length of binding peptides (13-25 amino acids). In the present study, we applied the Gibbs motif sampler, Sturniolo pocket profile and NNAlign method for binding motif identification and further position specific scoring matrices (PSSM) using sequence weighting schemes for the prediction of peptide binding to seven human MHC class-II molecules. Here, we used a non-parametric performance measure, area under receiver operating characteristic curve (Aroc) which provides a global assessment of predictive power. The average prediction performances for motif identification based on NNAlign, Sturniolo pocket profile and Gibbs sampler in term of Aroc are 0.71, 0.68 and 0.64, respectively. Further improvements in the performance of MHC class-II binding peptide predictor largely depends on the size of training dataset, optimization of training parameters and the correct identification of the peptide binding motifs.