Table 2. Comparison of the Predictive Performance of NetMHCpan-4.0_BA (the binding affinity prediction score of the NetMHCpan-4.0 method trained on both eluted ligand and peptide binding affinity data) and NetMHCpan-3.0 Models on Quantitative Binding Affinity Data from the IEDB Affinity Data Seta.
NetMHCpan-4.0_BA |
NetMHCpan-3.0 |
|||||
---|---|---|---|---|---|---|
BoLA-I | no. peps | no. bind | PCC | AUC | PCC | AUC |
BoLA-3*00101 (BoLA-AW10) | 166 | 8 | 0.497 | 0.816 | 0.381 | 0.792 |
BoLA-1*02301 (BoLA-D18.4) | 258 | 182 | 0.648 | 0.832 | 0.551 | 0.747 |
BoLA-6*01301 (BoLA-HD6) | 268 | 219 | 0.622 | 0.815 | 0.482 | 0.728 |
BoLA-3*00201 (BoLA-JSP.1) | 158 | 32 | 0.464 | 0.703 | 0.277 | 0.622 |
BoLA-T2c | 90 | 84 | 0.485 | 0.833 | 0.455 | 0.813 |
BoLA-2*01201 (BoLA-T2a) | 167 | 47 | 0.691 | 0.852 | 0.635 | 0.812 |
BoLA-6*04101 (BoLA-T2b) | 157 | 38 | 0.631 | 0.835 | 0.566 | 0.816 |
Ave | 0.577 | 0.812 | 0.478 | 0.761 |
Names in parentheses in the first column refer to the historical names for the different alleles. Performance was estimated in terms of Pearson’s correlation coefficient (PCC) and AUC (area under the receiver operator curve). Both of these performance measures take a value of 1 for the perfect and values of 0.0 (PCC)/0.5 (AUC) for a random prediction.