Table 5.
Predictive performance for the HLA ligand dataset EvaluationSet-3 dataset as measured in terms of the AUC.
| Allele | N | pan | ADT | KISS |
|---|---|---|---|---|
| A0201 | 104 | 0.955 | 0.925 | 0.948 |
| A0207 | 4 | 0.961 | 0.867 | 0.933 |
| A0301 | 5 | 0.986 | 0.977 | 0.987 |
| A1101 | 3 | 0.959 | 0.966 | 0.889 |
| A2402 | 2 | 0.998 | 0.988 | 0.998 |
| A2602 | 2 | 0.903 | 0.809 | 0.989 |
| A2603 | 1 | 0.807 | 0.898 | 0.977 |
| A3101 | 3 | 0.981 | 0.978 | 0.982 |
| A3301 | 2 | 0.918 | 0.798 | 0.632 |
| A6602 | 7 | 0.956 | 0.989 | 0.978 |
| A6603 | 2 | 0.999 | 0.999 | 0.996 |
| A6801 | 7 | 0.992 | 0.987 | 0.993 |
| A6802 | 1 | 0.999 | 0.999 | 0.994 |
| B0702 | 14 | 0.990 | 0.982 | 0.986 |
| B0801 | 20 | 0.986 | 0.829 | 0.962 |
| B0802 | 3 | 0.997 | 0.924 | 1.000 |
| B1509 | 1 | 0.866 | 0.842 | 1.000 |
| B1801 | 35 | 0.993 | 0.990 | 0.992 |
| B2702 | 2 | 0.989 | 0.987 | 0.996 |
| B2703 | 9 | 0.992 | 0.964 | 0.987 |
| B2704 | 23 | 0.985 | 0.951 | 0.984 |
| B2705 | 58 | 0.985 | 0.966 | 0.987 |
| B2706 | 25 | 0.980 | 0.945 | 0.980 |
| B2709 | 26 | 0.982 | 0.940 | 0.984 |
| B3901 | 61 | 0.967 | 0.563 | 0.884 |
| B4001 | 4 | 0.998 | 0.987 | 0.983 |
| B4101 | 12 | 0.978 | 0.915 | 0.689 |
| B4402 | 27 | 0.988 | 0.927 | 0.983 |
| B4403 | 1 | 0.970 | 0.967 | 0.997 |
| B4501 | 4 | 0.975 | 0.937 | 0.990 |
| B4701 | 18 | 0.909 | 0.604 | 0.789 |
| B4901 | 101 | 0.992 | 0.903 | 0.861 |
| B5001 | 8 | 0.990 | 0.955 | 0.801 |
| B5101 | 1 | 1.000 | 1.000 | 1.000 |
| Ave per allele | 34 | 0.968 | 0.919 | 0.945 |
| Ave per protein | 566 | 0.976 | 0.886 | 0.931 |
Allele gives the allele name. N gives the number of HLA ligands included in the benchmark for each allele. Pan (NetMHCpan-1.0), ADT and KISS give the AUC values averaged over all ligand:protein pairs for a given HLA allele for each of the three prediction methods included in the benchmark. Ave per allele gives the average of the per-allele performance values. Ave per protein gives the average over all 566 ligand:protein pairs included in the benchmark. For each allele, the best performing method is highlighted in bold.