Teleport‐stabilised quantum‐walk transport for robust neoantigen ranking in near‐tie regimes: A reproducible colorectal benchmark

Abstract Background Late‐stage neoantigen prioritisation often contains near ties among peptide–Human Leukocyte Antigen (HLA) candidates with similar presentation, processing, affinity and immunogenicity evidence. In a finite vaccine panel, unstable micro‐ordering can change which candidates cross the nomination boundary. Methods We performed an original computational study on an exact 56‐record colorectal peptide–HLA ledger with 19 source fields. Four prespecified continuous channels—presentation support, immunogenicity, transformed affinity and Transporter Associated with Antigen Processing (TAP) support—were normalised, embedded in an eight‐nearest‐neighbour Fubini–Study graph, propagated by classically evaluated continuous‐time quantum‐walk (CTQW) transport, stabilised by restart/teleport mixing and resolved by a deterministic modified black‐hole algorithm (MBHA) with non‐compensatory eligibility gates. Fixed weighted aggregation, binding‐only, immunogenicity‐only, reciprocal‐rank fusion (RRF), Borda aggregation and component controls were evaluated on the same rows. Rank stability was tested in 1000 matched seeded perturbation replicates. Results The declared presentation, immunogenicity and affinity gates retained 29 of 56 records. The full FS–CTQW–MBHA policy achieved median top‐10 retention of 0.90 (95% percentile interval 0.80–1.00) and median complete‐rank correlation of 0.999 (0.998–1.000). The fixed weighted aggregate achieved 1.00 and 0.998, respectively; binding‐only and immunogenicity‐only controls yielded 0.90/0.990 and 0.80/0.962. RRF yielded 0.80/0.927 and Borda 0.70/0.917. Thus, the full policy did not dominate every comparator on every stability metric. Conclusions The supported contribution is a stability‐first, graph‐aware decision layer that preserves very high complete‐rank reproducibility while adding non‐compensatory eligibility, redundancy governance, reason‐coded selected/reserve/failed states and deterministic replay. The study does not establish biological efficacy, clinical utility, candidate‐level Density Functional Theory/Quantum Monte Carlo (DFT/QMC) performance, biological quantum coherence or physical quantum‐computing advantage.