TAMAVAQ/Q-TAMAVAQ: Fubini–Study CTQW Graph Transport and Constrained Margin-Based Ranking for Neoantigen Candidate Prioritization in Lower-Grade Glioma and Glioblastoma

Personalized glioma neoantigen selection requires simultaneous control of mutation and expression evidence, peptide–HLA presentation, exact-sequence chemistry, structural plausibility, manufacturability, provenance, and finite-panel diversity under severe class imbalance. We present TAMAVAQ/Q-TAMAVAQ as a provenance-aware in silico peptide-selector architecture in which normalized candidate dossiers define a Fubini–Study similarity graph, continuous-time quantum-walk (CTQW) propagation supplies phase-sensitive spectral support, and constrained margin-based ranking assembles an HLA-balanced finite peptide panel. Across 7,262 source rows reconciled to 4,128 unique candidate records, together with a separate 1,537-record Seq2Neo benchmark containing 65 surrogate-positive peptide–HLA pairs (4.23%), the integrated framework achieved AP/AUPRC 0.552, AUROC 0.962, NDCG@100 0.693, TIF@25/50/100 of 0.680/0.660/0.450, and Recall@100 0.692. A matched classical control was evaluated on the identical Fubini–Study Laplacian using H(τ)=exp(−τL) with identical candidate identities, initialization, and downstream metrics. At the reference matched setting (k=8, σ equal to the median retained-neighbour dFS, t=τ=0.25), CTQW yielded AP/AUPRC 0.481 versus 0.471 for heat diffusion and NDCG@100 0.637 versus 0.632, while AUROC was 0.951 versus 0.956. Across the full 60-cell k × σ × time grid, CTQW exceeded heat diffusion in AP in 15 cells, defining a localized transport advantage pattern within a broader mixed response surface. Together, these results position CTQW as a complementary spectral neoantigen selector that sharpens a subset of near-tie ranking regimes within an auditable finite-panel workflow.