NN-align. An artificial neural network-based alignment algorithm for MHC class II peptide binding prediction

BMC Bioinformatics

Table 3 The HLA-DR ligand benchmark.

Allele	N	*NetMHCIIpan*	*TEPITOPE*	*NN-W-P1*	*NN-W*	*NN-xPFR*
DRB1*0101	37	0.873	0.883	0.899	0.882	0.863
DRB1*0301	26	0.882	0.837	0.862	0.906	0.788
DRB1*0401	209	0.865	0.876	0.865	0.843	0.848
DRB1*0404	46	0.817	0.790	0.776	0.772	0.770
DRB1*0405	35	0.848	0.809	0.892	0.866	0.878
DRB1*0701	36	0.687	0.711	0.761	0.754	0.757
DRB1*0802	1	0.982	0.914	0.984	0.979	0.959
DRB1*0901	4	0.865		0.867	0.864	0.878
DRB1*1101	27	0.873	0.863	0.894	0.876	0.881
DRB1*1302	21	0.605	0.761	0.702	0.687	0.681
DRB1*1501	12	0.770	0.729	0.767	0.766	0.776
DRB3*0101	2	0.957		0.680	0.730	0.681
DRB4*0101	4	0.471		0.540	0.492	0.496
DRB5*0101	15	0.840	0.853	0.877	0.819	0.851
Ave		0.810		0.812	0.804	0.794
Ave*		0.830		0.842	0.824	0.824
Ave**		0.822	0.821	0.844	0.834	0.824

The benchmark data set consists of 475 HLA-DR restricted ligands downloaded from the SYFPEITHI database of MHC ligands covering 14 HLA-DR alleles. The predictive performance was estimated in terms of the AUC as described in the text. Ave is the average per-allele performance over all 14 alleles. Ave* is the average predictive performance over all 475 ligand/HLA-DR pairs. Ave** is the average per allele performance over the 11 alleles covered by the TEPITOPE method. NetMHCIIpan is the HLA-DR pan-specific method described by Nielsen et al. [23]. TEPITOPE refers to the method developed by Sturniolo et al [17]. NN-W-P1 is the NN-based method including data redundancy step-size rescaling and PSSM-P1 amino acid encoding. NN-W is the NN-based method including data redundancy step-size rescaling. NN-xPFR is the NN-W-P1 method excluding peptide flanking residue encoding. For each allele, the best performing NN method is highlighted in bold and the best performing of all methods is underlined.

ISSN: 1471-2105