Improved methods for predicting peptide binding affinity to MHC class II molecules

Abstract

Major histocompatibility complex class II (MHC-II) molecules are expressed on the surface of professional antigen-presenting cells where they display peptides to T helper cells, which orchestrate the onset and outcome of many host immune responses. Understanding which peptides will be presented by the MHC-II molecule is therefore important for understanding the activation of T helper cells and can be used to identify T-cell epitopes. We here present updated versions of two MHC-II-peptide binding affinity prediction methods, NetMHCII and NetMHCIIpan. These were constructed using an extended data set of quantitative MHC-peptide binding affinity data obtained from the Immune Epitope Database covering HLA-DR, HLA-DQ, HLA-DP and H-2 mouse molecules. We show that training with this extended data set improved the performance for peptide binding predictions for both methods. Both methods are publicly available at www.cbs.dtu.dk/services/NetMHCII-2.3 and www.cbs.dtu.dk/services/NetMHCIIpan-3.2.

Keywords: MHC binding specificity; T-cell epitope; affinity predictions; immunogenic peptides; peptide-MHC binding.

Figure 1

Performance of NetMHCII‐2.3 and NetMHCIIpan‐3.2 together with the combination method. (a) The average performance per MHC molecule of NetMHCII‐2.3, NetMHCIIpan‐3.2 and the combination method, including the significance between the methods. P‐values where found using a paired t‐test using the predictions per molecule found in Table S3 (see the Supplementary material). (b) The average predictive performance of the MHC molecules in the data set as a function of the number of peptides. [Colour figure can be viewed at http://wileyonlinelibrary.com]

Figure 2

Predictive performance for NetMHCIIpan‐3.2 LOMO on the MHC class II molecules from the 2016 data set as a function of distance to the nearest neighbour. Each HLA II isotype and H‐2 molecules are displayed in different colours and the dashed line represents the least square fit for the data. [Colour figure can be viewed at http://wileyonlinelibrary.com]

Figure 3

Distance tree for all HLA molecules found in our data set generated using the MHCC luster method. Sequence logos shows the motif of the predicted binding core for each HLA and were generated using Seq2Logo.30 [Colour figure can be viewed at http://wileyonlinelibrary.com]

Figure 4

Performance of NetMHCIIpan‐3.1 and NetMHCIIpan‐3.2 using the T‐cell epitope benchmark set. (a) The average Frank performance per MHC molecule for the two versions of NetMHCIIpan. (b) The average AUC performance per MHC molecule for the two versions of NetMHCIIpan. (c) The change in the distance to the nearest neighbour between the two data sets used for training the old and the new versions of NetMHCIIpan as a function of the change in distance to the nearest neighbour. (d) the change in the number of data points between the two data sets used for training NetMHCIIpan‐3.1 and NetMHCIIpan‐3.2 as a function of the change in the performance, including only MHC molecules where the pseudo sequence did not change between two data sets. The dashed line in the two scatterplots represents the least square fit for the data.