NetMHCIIpan-2.0 - Improved pan-specific HLA-DR predictions using a novel concurrent alignment and weight optimization training procedure

Table 8 Endogenous HLA-DR ligand benchmark

SYF	#	*NetMHCIIpan-1.0*	*NetMHCIIpan-2.0*	*TEPITOPE*
Ave per ligand	1164	0.800	0.829
Ave per allele	28	0.788	0.797
In TEPITOPE	17	0.768	0.786	0.799
!In TEPITOPE	11	0.819	0.814
IEDB	#	*NetMHCIIpan-1.0*	*NetMHCIIpan-2.0*	*TEPITOPE*
Ave per epitope	1325	0.729	0.751
Ave per allele	42	0.759	0.781
In TEPITOPE	20	0.745	0.747	0.755
!In TEPITOPE	22	0.772	0.812

NetMHCIIpan-1.0 is the method described by Nielsen et al. [29], NetMHCIIpan-2.0 is the pan-specific method described here, and TEPITOPE is the method described by Sturniolo et al. [1]. Ave per ligand/epitope gives the average AUC over the 1164/1325 ligands/epitopes in the benchmark data set. Ave per allele gives the average over the per allele averaged AUC values. In TEPITOPE gives the per allele average of the subset of alleles characterized by the TEPITOPE method, and !In TEPITOPE give the per-allele average performance of the alleles not characterized by the TEPITOPE method. AUC values were calculated as described in the text. For each benchmark subset, the best performing method is highlighted in bold.