Evaluation of immunoproteasome and proteasome prediction models. The predictive performance of proteasome models was evaluated in 5 fold cross-validation experiments using MCC, BTR, SP and SE as measures of performance. Proteasome models were built and tested using MHCI-eluted peptide ligands whereas immunoproteasome models were built and tested using MHCI-restricted CD8 T cell epitopes. A) Predictive performance (BTR) achieved by immunoproteasome (grey line) and proteasome (black line) models trained and tested on peptides of different fragment lengths (abscissa). Statistically significant increments in BTR are indicated with "*" symbols. B) Predictive performance (BTR, MCC, SP, SE) achieved by the selected proteasome and immunoproteasome models built on peptide fragments of 12 residues.