Abstract
Proteogenomic cancer antigen discovery now extends beyond somatic mutations to include unmutated, aberrantly expressed, and non-canonical sources. We distinguish two orthogonal dimensions: molecular origin (canonical or non-canonical) and normal-tissue distribution (tumor-specific antigen, tumor-associated antigen, or lineage-specific antigen). Importantly, non-canonical origin does not imply tumor specificity. For human leukocyte antigen (HLA)-restricted immunotherapy, selectivity must be established at the level of naturally presented peptide-HLA (pHLA) complexes rather than inferred from RNA expression or total protein abundance. We therefore propose a normal-immunopeptidome denominator: a context-aware atlas of benign pHLA ligands that can be used to exclude unsafe candidates before therapeutic nomination. However, this denominator is necessary but not sufficient. Candidate nomination also requires evidence of analytical confidence, natural presentation on tumor cells, and, where available, peptide abundance and tumor heterogeneity. Additional considerations include whether the target is patient-specific or shared, its functional selectivity, the potential for immune escape, receptor cross-reactivity, HLA alloreactivity, and the modality-specific therapeutic index. We present a tiered framework that distinguishes core requirements, risk-triggered evaluation criteria, and recommended enhancements. These requirements escalate as uncertainty increases. Importantly, the absence of a peptide from existing databases should not be interpreted as evidence of its absence in humans.