Table 4.
Five-fold cross-validation performance of the pan-specific NetMHCIIpan-2.0 method compared to the allele-specific NN-align and TEPITOPE methods on the quantitative benchmark data set
| NN-align | NetMHCIIpan-2.0 | TEPITOPE | |||||
|---|---|---|---|---|---|---|---|
| Allele | # | # bind | PCC | AUC | PCC | AUC | AUC |
| DRB1*0101 | 7685 | 4382 | 0.675 | 0.825 | 0.711 | 0.846 | 0.727 |
| DRB1*0301 | 2505 | 649 | 0.690 | 0.855 | 0.709 | 0.864 | 0.718 |
| DRB1*0302 | 148 | 44 | 0.272 | 0.659 | 0.525 | 0.757 | |
| DRB1*0401 | 3116 | 1039 | 0.643 | 0.833 | 0.670 | 0.848 | 0.762 |
| DRB1*0404 | 577 | 336 | 0.565 | 0.766 | 0.630 | 0.818 | 0.747 |
| DRB1*0405 | 1582 | 627 | 0.710 | 0.869 | 0.698 | 0.858 | 0.780 |
| DRB1*0701 | 1745 | 849 | 0.718 | 0.855 | 0.740 | 0.864 | 0.777 |
| DRB1*0802 | 1520 | 431 | 0.518 | 0.778 | 0.526 | 0.780 | 0.645 |
| DRB1*0806 | 118 | 91 | 0.744 | 0.902 | 0.796 | 0.924 | 0.884 |
| DRB1*0813 | 1370 | 455 | 0.729 | 0.878 | 0.746 | 0.885 | 0.750 |
| DRB1*0819 | 116 | 54 | 0.370 | 0.706 | 0.608 | 0.808 | |
| DRB1*0901 | 1520 | 622 | 0.597 | 0.810 | 0.634 | 0.818 | |
| DRB1*1101 | 1794 | 778 | 0.756 | 0.873 | 0.777 | 0.883 | 0.793 |
| DRB1*1201 | 117 | 81 | 0.699 | 0.860 | 0.764 | 0.892 | |
| DRB1*1202 | 117 | 79 | 0.695 | 0.866 | 0.769 | 0.900 | |
| DRB1*1302 | 1580 | 493 | 0.630 | 0.819 | 0.634 | 0.825 | 0.596 |
| DRB1*1402 | 118 | 78 | 0.623 | 0.825 | 0.694 | 0.860 | |
| DRB1*1404 | 30 | 16 | 0.466 | 0.661 | 0.613 | 0.737 | |
| DRB1*1412 | 116 | 63 | 0.680 | 0.857 | 0.757 | 0.894 | |
| DRB1*1501 | 1769 | 709 | 0.641 | 0.815 | 0.653 | 0.819 | 0.731 |
| DRB3*0101 | 1501 | 281 | 0.673 | 0.843 | 0.690 | 0.850 | |
| DRB3*0301 | 160 | 70 | 0.604 | 0.826 | 0.736 | 0.853 | |
| DRB4*0101 | 1521 | 485 | 0.701 | 0.854 | 0.675 | 0.837 | |
| DRB5*0101 | 3106 | 1280 | 0.735 | 0.865 | 0.765 | 0.882 | 0.760 |
| Ave | 33931 | 13992 | 0.631 | 0.821 | 0.688 | 0.846 | |
| Ave* | 0.673 | 0.841 | 0.697 | 0.854 | 0.744 | ||
| Ave** | 0.580 | 0.797 | 0.679 | 0.837 | |||
# gives the number of peptide binding data for each allele, #bind gives the number of peptides with a binding affinity stronger than 500 nM. NN-align is the method described by Nielsen et al. [5], NetMHCIIpan-2.0 is the method described here, and TEPITOPE is the method described by Sturniolo et al. [1]. Ave gives the per allele average, Ave* gives the per allele average of the 13 alleles characterized by the TEPITOPE method, and Ave** gives the per allele average of the 11 allele not characterized by the TEPITOPE method. In bold is highlighted the best performing method for each of the 24 alleles. AUC values were calculated using a binding threshold of 500 nM. Only AUC values are included for the TEPITOPE method since prediction values for this method are not linearly related to the binding affinity.