Benchmarking a mechanistic immunogenicity framework: aligning sequence-based epitope predictions with clinical observations