AbLWR: A Context-Aware Listwise Ranking Framework for Antibody-Antigen Binding Affinity Prediction via Positive-Unlabeled Learning.

AbLWR reformulates antibody–antigen binding affinity prediction as listwise ranking and improves Full Rank Accuracy by approximately 8% versus the strongest baseline in randomized cross-validation.

• Why it matters: Point-wise regression struggles to separate subtle affinity differences among highly homologous variants and cannot leverage abundant unlabeled interactions under severe label sparsity.
• What they did: AbLWR uses Positive-Unlabeled learning with a dual-level contrastive objective and meta-optimized label refinement, plus homologous antigen sampling with an MHSA module to model dependencies within training lists.
• The result: In randomized cross-validation, it raises FRA by ~8% and shows robustness in out-of-distribution settings with distinct antigen and antibody clusters, validated by case studies on Influenza viruses and Human Interleukin-33.