Computational vaccinology is a developing discipline. To become a standard component in vaccine development, it requires accurate and broadly applicable models of wet-lab experiments. We developed prediction models based on a novel data representation of peptide/MHC interaction and support vector machines (SVM) for prediction of peptides that promiscuously bind to multiple Human Leukocyte Antigen (HLA) alleles belonging to HLA-B7 supertype. 10-fold cross-validation results showed that the area under the receiver operating curve (Aroc) of SVM models is above 0.90. Blind testing results showed that the average Aroc of SVM models is 0.84. A learning approach based on information theory, termed Information Learning Approach, was used for feature selection. Several amino acid positions with high information content have been identified in input 9mer peptides and HLA alleles and were used as input features to SVM. They are position 1, 2, 4, 5, 7, 8, 9 in 9mer peptides and position 45 and 97 in HLA-B7 molecules. Prediction accuracy was improved after feature selection. These positions cover the anchor positions of HLA-B7 alleles, which have important biological roles for successful biding of relevant peptides.