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

2026-05-20

AI Summary

This work addresses a key limitation in traditional antibody CDR design methods, which fail to distinguish between antigen contact recognition and amino acid selection, thereby diluting antigenic signals uniformly across all positions. To overcome this, the authors propose ConTact, a novel architecture that explicitly models the contact inference process by decomposing CDR design into three cascaded stages: learning surface complementarity fingerprints, predicting CDR–antigen contacts, and generating sequences via contact-gated decoding. The method incorporates distance-biased cross-attention to encode spatial proximity priors and employs a contact-weighted loss function to emphasize critical binding residues. Evaluated on CHIMERA-Bench, ConTact achieves state-of-the-art structural accuracy (7% improvement in RMSD), superior epitope awareness (10% higher F1 score than GNN baselines), and competitive sequence recovery (AAR of 0.38).

Abstract

Computational antibody CDR design methods condition on antigen structure to generate binding loops, yet existing architectures conflate two fundamentally distinct sub-problems: identifying which CDR positions will contact the antigen, and selecting amino acids at those positions. This conflation 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 sequence 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 CHIMERA-Bench dataset, ConTact achieves the best structural quality (7% RMSD improvement over the next-best baseline), best epitope awareness (10% F1 score over GNN baselines), and competitive sequence recovery (AAR 0.38) among several CDR-H3 design baselines.

Problem

• antibody CDR design
• antigen contact prediction
• binding interface
• sequence-structure modeling
• epitope recognition

Innovation

• contact-first design
• explicit interface reasoning
• distance-biased cross-attention
• contact-weighted loss
• CDR-H3 design