NetMHCpan, a method for MHC class I binding prediction beyond humans

Abstract

Binding of peptides to major histocompatibility complex (MHC) molecules is the single most selective step in the recognition of pathogens by the cellular immune system. The human MHC genomic region (called HLA) is extremely polymorphic comprising several thousand alleles, each encoding a distinct MHC molecule. The potentially unique specificity of the majority of HLA alleles that have been identified to date remains uncharacterized. Likewise, only a limited number of chimpanzee and rhesus macaque MHC class I molecules have been characterized experimentally. Here, we present NetMHCpan-2.0, a method that generates quantitative predictions of the affinity of any peptide-MHC class I interaction. NetMHCpan-2.0 has been trained on the hitherto largest set of quantitative MHC binding data available, covering HLA-A and HLA-B, as well as chimpanzee, rhesus macaque, gorilla, and mouse MHC class I molecules. We show that the NetMHCpan-2.0 method can accurately predict binding to uncharacterized HLA molecules, including HLA-C and HLA-G. Moreover, NetMHCpan-2.0 is demonstrated to accurately predict peptide binding to chimpanzee and macaque MHC class I molecules. The power of NetMHCpan-2.0 to guide immunologists in interpreting cellular immune responses in large out-bred populations is demonstrated. Further, we used NetMHCpan-2.0 to predict potential binding peptides for the pig MHC class I molecule SLA-1*0401. Ninety-three percent of the predicted peptides were demonstrated to bind stronger than 500 nM. The high performance of NetMHCpan-2.0 for non-human primates documents the method's ability to provide broad allelic coverage also beyond human MHC molecules. The method is available at http://www.cbs.dtu.dk/services/NetMHCpan.

Fig. 1

Average performance of NetMHCpan-1.0 and NetMHCpan-2.0 on HLA-A and HLA-B molecules. The performance is given as Pearson's correlation coefficient. The significance of the difference in performance for HLA-B was tested using a paired one-tailed t test (n=29)

Fig. 2

Binding motifs of Mamu-A*02, Mamu-A*11, Patr-A*0101, and Patr-B*1301 generated from reported and predicted binders. The predicted binders were generated as the top scoring 1% best binders of 100,000 randomly selected natural 9mer peptides. Position specific scoring matrices (PSSM) were calculated from the set of binding peptides using sequence weighting and correction for low counts (Altschul et al. 1997; Nielsen et al. 2004). The binding motifs were visualized using the logo-plot method by Schneider and Stephens (1990). In a sequence logo, the height of a column of letters is equal to the information content at that position, and the height of each letter within a column is proportional to the frequency of the corresponding amino acid at that position

Fig. 3

Estimation of the NetMHCpan-2.0 prediction performance. The graph shows the result of a fivefold cross-validation (n=82). The Pearson's correlation coefficient (PCC) between observed and predicted performance is 0.67 (R²=0.45)

Fig. 4

Fivefold cross-validation performances of NetMHCpan-2.0. The histogram shows the average Pearson's correlation coefficient for HLA-A, HLA-B, rhesus macaque (Mamu), chimpanzee (Patr), and mouse MHC class I molecules

Fig. 5

Scatter plots of the predicted versus experimental IC50 values for the HLA-A*0302 alleles. NetMHCpan refers to the method developed in this paper, and NetMHC refers to the single-allele neural-network-based method developed by Lundegaard et al. (2008). The lines in the plots are least square fits for NetMHCpan (solid line) and NetMHC (dashed line), respectively. The HLA-A*0302 is characterized with 21 peptide data points. The Pearson's correlation between the prediction and experimental log(IC50) values is 0.77 and 0.29 for the NetMHCpan and NetMHC methods, respectively, and the slope of the best linear fit is 0.71 and 4.04

Fig. 6

Predicted binding motifs of HLA-Cw*0102 and Cw*0304 and the reported ligands (Rammensee et al. 1999). The motif sequence logos were generated as described in Fig. 3

Fig. 7

Predicted binding motif of HLA-G*0101 and reported ligands (Rammensee et al. 1999). The motif sequence logos were generated as described in Fig. 3

Fig. 8

Sequence motifs generated from the 13 verified binders and predicted binders for the swine MHC class I molecule SLA-1*0401. The predicted binders and motif sequence logos were generated as described in Fig. 3