Modelling and predicting antigen presentation with Restricted Boltzmann Machines

Immune recognition of infected and malignant cells requires presentation on their surface of antigens (i.e. short peptides) by human leukocyte antigen class I (HLA-I) proteins, which are coded by one of the most polymorphic alleles in the human genome. The identification of clinically relevant, tumour-specific neoantigens (mutated antigens) is currently a highly sought-after goal in designing novel cancer immunotherapeutic strategies. Algorithms aimed at predicting peptide presentation by HLA-I proteins are therefore valuable tools to accelerate the validation of putative neoantigens. To tackle this problem, we resort to a framework of inference from aminoacid sequences based on Restricted Boltzmann Machines, probabilistic graphical models characterized by a layer of ‘feature’ variables. This approach ensures efficient prediction of what antigens can be presented along with their HLA-I binding specificity; furthermore, it can be used to study in a model-guided way the effect of mutations on antigen presentation.