Identification of MHC ligands through allele-guided isolation combined with machine learning for improved MHC assignment using ARDisplay-I

The isolation of MHC ligands and subsequent analysis by mass spectrometry is considered the gold standard for defining targets for T cell-based immunotherapies. However, as many targets of high tumor specificity are only presented at low abundance on the cell surface of tumor cells, the efficient isolation of these peptides is crucial for their successful detection. Here, we demonstrate how optimizing the MHC ligand isolation strategy, based on both the presenting MHC alleles and the individual peptide level, enhances the identification of specific MHC ligands. This ideally acknowledges not only the hydrophobicity but also the post-translational modifications of the respective MHC ligands. To further improve the identification and characterization of MHC ligands, we developed an MHC class I ligand prediction algorithm (ARDisplay-I) that outperforms current state-of-the-art tools when benchmarked against competitors such as netMHCpan 4.1, MixMHCpred, or MHCflurry. Implementing these strategies can augment the development of T cell receptor-based therapies by improving the identification of novel immunotherapy targets and enriching the resources available in the computational immunology field through a superior MHC presentation prediction algorithm. SIGNIFICANCE: The multiallelic character of almost all samples in immunopeptidomics adds to the complexity of MHC ligand isolation and assignment. This study demonstrates how the isolation of different MHC ligands can be optimized when considering their hydrophobicity and post-translational modification status. Additionally, to improve the assignments of these MHC ligands to their respective MHC alleles in a multiallelic setting, we developed a machine-learning model to predict the probability of presentation of these MHC ligands on the cell surface, which was successfully benchmarked against widely used algorithms. Both approaches, especially if combined, have the potential to improve the detection of targets for the design of pHLA-targeted immunotherapies.