Robustness analysis. The reproducibility of the results of the prediction methods and their dependence on the size of the available dataset was examined in selected alleles. Box plots of randomly selected balanced sets of binders and non-binders from Dataset F for the alleles A*0201, A*3101, and B*0702 are shown. The smallest dataset for each allele consisted of 50 binders and 50 non-binders. The size of the largest dataset for each allele depends on the total number of binders or non-binders available for that particular allele. NetMHC was not included in this analysis as the predictor is only available online and could therefore not be trained by the authors.