Benchmarking Experimental and AlphaFold 3 Scaffolds for $\Delta\Delta G$-Based TCR Mutation Prioritization

Machine learning-based binding affinity prediction can support T-cell receptor (TCR) engineering, but its reliability remains uncertain when experimental structures of TCRs bound to peptide-major histocompatibility complexes (pMHCs) are unavailable and predicted structures must be used instead. In this study, we examined how AlphaFold 3 (AF3) scaffold choice and conformational sampling affect $\Delta\Delta G$-based prioritization of TCR-pMHC complementarity-determining region (CDR) mutations. Comparison of bound and unbound structures showed that pMHC binding is associated mainly with local CDR remodeling, especially in CDR3 loops. Across four structure-based $\Delta\Delta G$ prediction models, performance depended strongly on scaffold source and model architecture. Experimental structures remained the most reliable structural inputs when available. When experimental structures were unavailable, mutant-specific AF3 scaffolds were useful for recovering top-ranked CDR candidates across the evaluated templates.