ConTact: Contact-First Antibody CDR Design via Explicit Interface Reasoning

Computational antibody CDR design methods condition on antigen structure to generate binding loops. Yet, the existing architectures conflate two fundamentally distinct sub-problems: identifying which CDR positions will contact the antigen, and selecting amino acids at those positions. This forces models to learn contact reasoning implicitly through uniform message passing, diluting antigen signal across all positions equally. We introduce ConTact, a contact-then-act architecture that explicitly decomposes CDR design into three cascaded stages: learning surface complementarity fingerprints, predicting CDR-antigen contacts, and injecting contact-gated antigen features into the prediction head. A distance-biased cross-attention module encodes geometric priors favoring spatial neighbors, while a contact-weighted cross-entropy loss concentrates gradient signal on binding-critical positions. On the CHIMERA-Bench dataset, ConTact achieves the lowest backbone RMSD on every split (a 5 to 6% improvement over the best baseline) and the best fraction of native contacts, interface RMSD, and epitope F1 on the antigen-fold and temporal splits, while remaining competitive on the harder epitope-group split. The source code is available at: https://github.com/mansoor181/ConTact.git