Footprints of antigen processing boost MHC class II natural ligand predictions

Table 4 Independent evaluation of eluted ligand data set in terms of AUC 0.1

Training set		Eval set	BA single	BA combined	EL single	EL combined	NetMHCIIpan	NetMHCII
BA	EL	Eval set	BA single	BA combined	EL single	EL combined	NetMHCIIpan	NetMHCII
DR1 BA	DR1 Ph	DR1 Pm	0.57	0.644	0.839	0.849	0.562	0.585
	DR1 Ph	DR1 Sm	0.573	0.648	0.81	0.813	0.576	0.578
	DR1 Pm	DR1 Ph	0.498	0.557	0.757	0.754	0.478	0.469
	DR1 Pm	DR1 Sm	0.514	0.549	0.665	0.669	0.519	0.511
	DR1 Sm	DR1 Ph	0.506	0.556	0.704	0.707	0.481	0.478
	DR1 Sm	DR1 Pm	0.568	0.636	0.824	0.835	0.54	0.57
DR15 BA	DR15 Ph	DR15 Pm	0.584	0.672	0.869	0.869	0.427	0.58
DR15 BA	DR15 Pm	DR15 Ph	0.615	0.71	0.888	0.889	0.459	0.583

“Training set” refers to the data sets used to train the given model (BA indicates binding affinity and EL eluted ligand data). For references on the training sets names and compositions, refer to Table 1. Cross-validated performance values are reported as AUC 0.1. “Eval set” refers to the independent eluted ligand data set from the same allotype used for evaluation. “BA single” or “EL single” refers to model trained on single data types (BA or EL respectively). “BA combined” or “EL combined” refers to the eluted ligand prediction output or binding affinity output of models trained on both data types. NetMHCIIpan or NetMHCII refers to predictions made using the NetMHCIIpan 3.2 [36], and NetMHCII 2.3 [36] publicly available prediction methods

ISSN: 1756-994X