Research questionHow can antibody–antigen binding affinity be predicted from sequences without resolved three-dimensional structures?Resolved antibody–antigen structures are costly and scarce, yet screening requires estimating binding strength across many sequence pairs. Sequence-only prediction must capture interactions among antigen and antibody-chain sequences without direct structural information. Latest papersRecent research connected to this question, newest first.DuaDeep-SeqAffinity: Dual-Branch Deep Learning for Tri-Stream Sequence-Based Antibody--Antigen Affinity PredictionThe source presents DuaDeep-SeqAffinity, which processes antigen, heavy-chain, and light-chain sequences as separate streams using frozen ESM-2 embeddings, Transformer and CNN branches, and late fusion. On a sequence-disjoint AbRank split, it reports Pearson correlation of 0.683, R² of 0.460, and pairwise ranking AUC of 0.895, outperforming single-branch ablations; saliency analyses emphasize CDR loops and candidate epitope residues. Evidence is limited to this benchmark and reported analyses.research paper · Sep 3, 2026