Immunogenic response prediction techniques are described. In an example, a
system receives first training data that identifies
a peptide, a first set of multiple
major histocompatibility complex (MHC) molecules, and an immunogenic response associated with the
peptide and the first set. The
system updates a parameter of a first model based at least in part on the first training data. The first model is configured to determine a probability of causing the immunogenic response by a pair formed by the
peptide and a MHC molecule from the first set. The
system also receives second data that identifies a second set of MHC molecules, and generates, by using the first model and a second model, an immunologic response prediction of
pairing the
peptide with a MHC molecule from the second set. The second model is configured to generate at least one of peptide-MHC binding predictions or peptide-MHC
cell surface presentation predictions.