Clinical correlation of AI neoantigen prediction models

Establish whether AI-based neoantigen prediction models, specifically DeepNeoAG and ImmuneMirror, exhibit statistically significant correlations between their predicted immunogenicity scores and clinical outcomes such as overall survival, progression-free survival, recurrence-free survival, or relapse rates in melanoma cohorts and prospective trials, thereby confirming translational validity beyond in vitro peptide–MHC binding performance.

Background

The paper introduces the Algorithm-to-Outcome Concordance (AOC) metric to bridge AI model performance with clinical outcomes in neoantigen vaccine development. Despite strong in vitro results for peptide–MHC binding prediction, the authors note a critical translational gap: existing AI frameworks lack established correlations with real patient outcomes, making clinical utility uncertain.

Resolving this gap requires prospective validation linking model-derived immunogenicity scores to patient-level endpoints. Demonstrating such correlations is essential for regulatory adoption and for ensuring AI-driven vaccine design yields tangible clinical benefits.