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

BMC Bioinformatics

Table 5 Predictive performance in terms of the AUC on the Lin benchmark data set.

Method	DRB1*0101	DRB1*0301	DRB1*0401	DRB1*0701	DRB1*1101	DRB1*1501	Ave
IEDB_SMM	0.81	0.71	0.79	0.67	0.84	0.67	0.75
IEDB_SAT	0.89	0.69	0.75	0.74	0.83	0.66	0.76
IEDB_Cons	0.83	0.67	0.72	0.80	0.84	0.66	0.75
Multipred_SVM	0.86	0.80	0.65	0.70	0.78	0.62	0.74
NetMHCII	0.77	0.69	0.81	0.62	0.78	0.65	0.72
NetMHCIIpan	0.84	0.65	0.81	0.83	0.81	0.79	0.79
Propred	0.89	0.70	0.75	0.74	0.83	0.66	0.76
SVMHC	0.86	0.69	0.75	0.74	0.83	0.66	0.76
SYF	0.72	0.65	0.69	0.70	0.59	0.77	0.69
NN-W-P1	0.90	0.78	0.84	0.75	0.85	0.79	0.82
NN-W	0.90	0.79	0.80	0.76	0.83	0.82	0.81

NN-W-P1 is the NN-based method including data redundancy step-size rescaling and P1-PSSM encoding and NN-W the NN-based method including data redundancy step-size rescaling. All other methods are described in the Lin et al. publication. The AUC was calculated using the following binding affinity threshold values for each of the 6 alleles: DRB1*0101, 0401, 0701, and 1501 threshold = 100 nM, DRB1*0301, 1101, and 1301, threshold = 1000 nM (Lin HH personal communication). The performance values for the 9 methods above the dotted line were taken from Lin et al. [24]. For each allele, the best performing NN method is highlighted in bold and the best performing method is underlined.

ISSN: 1471-2105