Structure-Informed Scoring Bridges the Generation–Selection Gap in Antibody–Antigen Structure Prediction

Protein co-folding models often sample accurate antibody–antigen complex structures, yet their confidence scores fail to reliably identify them, exposing a gap between structure generation and candidate selection. In particular, original confidence is more sensitive to variation across seeds than to structural differences among diffusion samples within a seed. Extending prior work on structure-informed rescoring, we introduce ReCon, which reconstructs a candidate-specific representation from each full-complex structure while preserving its interchain geometry for confidence evaluation. By scoring candidates with separately reconstructed representations, ReCon improves overall ranking and exhibits stronger discrimination among samples generated within the same seed. Uncertainty in the original pair representations further identifies targets likely to benefit from rescoring, enabling selective application at reduced computational cost. Together, these results establish full-complex, candidate-conditioned representation reconstruction as a practical, uncertainty-guided approach to improving antibody–antigen structure selection.