Prediction of epitope immunogenicity for CD8+ Cell
A high immunogenicity score indicates increased confidence that the denoted peptide-HLA complex will trigger a T cell response, compared to another peptide with a lower immunogenicity score. However, it does not necessarily reflect the actual degree of immunogenicity, due to a lack of TCR sequence diversity information for a given individual. Specifically, the immunogenicity score reflects confidence based on experimental evidence of for immunogenicity in the training data. For example, if a single peptide (peptide-1) produced an immune response in 40 out of 40 experiments, it will result in a higher confidence observation than another peptide (peptide-2) that results in immune response 1 out of 6 experiments. DeepImmunoCNN uses a beta-binomial model to incorporate this evidence when deriving an immunogenicity potential from the training data. Hence, the output value reflects the confidence of the prediction.
Selecting this option will additionally produce peptide-MHC binding predications in addition to immunogenicity predictions for queried peptide(s). Any valid CD8+ T cell mediated immune response requires a peptide is presented by the HLA allele and the HLA-peptide complex triggers a T-cell response. DeepImmuno-CNN predicts immunogenicity and not MHC-binding. By selecting this option, the user can assess the likelihood of both binding and immunogenicity in a single query. As such, this analysis requires approximately 30-60s additional runtime, as it calls the python package MHCflurry.
In certain cases, the peptide query for a specific HLA allele, will produce unreliable results, when the HLA-allele has few immunogenicity experimental results in our training dataset. As such, DeepImmunno-CNN is not able to reliably infer its immunogenicity.