Comparative results of DistBoost and RANKPEP on the H-2Kd MHC class I molecule. The left plot presents ROC (see Evaluation methods section for details) curves of the best test score obtained when training on 50% of the entire data (red: using only positive constraints; blue: using both types of constraints). The intersection between the curves and the diagonal line marks the equal error-rate statistic. The right plot presents average AUC scores on test data. We compare the two PSSM methods used by RANKPEP (A: PROFILEWEIGHT, B: BLK2PSSM) to DistBoost when trained using only positive constraints (C) and when trained using both positive and negative constraints (D). The averages were taken over 10 different runs on randomly selected train and test sets. N denotes the total number of binding peptides (of which 50% were used in the training phase and the remaining 50% were used in the test phase). For a detailed comparison see Figs. 5-6.