NetMHCpanExp - 1.0

Pan-specific binding of peptides to MHC class I proteins of known sequence including antigen expression

The NetMHCpanExp-1.0 server predicts binding of peptides to any human MHC-I molecule of known sequence using artificial neural networks (ANNs). This NetMHCpan-like method incorporates a new feature in the prediction of peptide-MHC class I binding: gene expression.

The method is trained on a combination of more than 670,000 quantitative Binding Affinity (BA) and Mass-Spectrometry Eluted Ligands (MS EL) peptides. The BA data covers 112 human MHC-I molecules (HLA-A, B, C and E) while the EL data covers 163 human MHC-I molecules (HLA-A, B, C and G). The user can obtain predictions to any custom human MHC class I molecule by uploading a full length MHC protein sequence. Predictions can be made for peptides of length 8 to 14.

This method, as well as its closest predecessor NetMHCpan-4.1, is built upon the algorithm NNAlign_MA which allows for simultaneous training and pseudo-labelling (automatic annotation) of multi-allelic data (MA), derived from MS experiments performed on cell lines or tissue samples. Additionally, in this case, the architecture of the algorithm NNAlign_MA was modified to accept gene expression values in the peptide encoding.

The current method was trained with MHC-I ligands derived from samples that were also assayed in RNA-Seq experiments. Finally, the training set was enlarged including MS EL ligands that originally lacked gene expression values by use of transcript abundance values derived from RNA-Seq experiments performed on 281 human tissue and blood cell samples deposited in the Human Protein Atlas (v. 20.0) database.

The user can make predictions for peptide (or protein) sequences with already annotated gene expression values (in Transcripts Per Millon). If the gene expression values are absent from the input data, the tool allows for automatic annotation of these values employing the Human Protein Atlas (v. 20.0) reference database. For more details, click on Instructions. Also, the user is enabled to make predictions without including the novel feature.

The server returns the likelihood of a peptide being a natural ligand of the selected MHC(s).

The project is a collaboration between the Bioinformatics Section at DTU HealthTech (see all available services) and LIAI.


Correspondence:        Technical Support: