ABAG-Rank: improving model selection of AlphaFold antibody-antigen complexes by learning to rank.

ABAG-Rank improves antibody-antigen complex model selection by outperforming AlphaFold3's internal scores, achieving a significant ranking performance boost.

• Why it matters: Accurate prediction of antibody-antigen interactions is crucial for vaccine and therapeutic development, but current methods struggle to reliably distinguish correct models without strong co-evolutionary signals.
• What they did: The team developed ABAG-Rank, a deep neural network based on DeepSets architecture, trained on known complexes, using geometric descriptors and AlphaFold confidence scores to rank structural decoys effectively.
• The result: ABAG-Rank surpasses AlphaFold3 internal scoring and existing deep learning methods, enabling more accurate identification of correct antibody-antigen models and facilitating advances in immunological research.