Immunogenicity
How strongly something provokes an immune response — the property that separates a neoantigen that works from one that's merely present.
A peptide can be perfectly presented on HLA and still fail to trigger T cells. Immunogenicity is the actual capacity to elicit a response, and it's the property vaccine developers ultimately care about.
Because immunogenicity depends on the patient's TCR repertoire and many subtle factors, predicting it remains a leading challenge — and a key place where better AI models could materially improve vaccine response rates.
Can AI predict which neoantigens are immunogenic?
Partly. AI predicts HLA presentation well, but immunogenicity — whether a T cell actually responds — is much harder, because it also depends on the patient's individual T-cell receptor repertoire and on factors that current datasets capture poorly. Improving immunogenicity prediction is the field's most active AI research front.
Why is immunogenicity prediction so difficult?
A peptide can be perfectly presented and still provoke no response. Immunogenicity depends on whether the patient happens to have a T-cell receptor that recognizes the peptide–HLA complex, plus self-tolerance and the tumor microenvironment — variables that are individual, sparsely measured, and not fully understood.