Relation-Aware Graph-Hypergraph Networks for Antibody Sequence-Structure Co-Design

Abstract

Motivation: The design of antibody complementarity-determining regions (CDRs) is pivotal for therapeutic development, yet remains challenged by the complexity of the antibody-antigen interface. Existing methods, which largely rely on pairwise relationships, struggle to capture intrinsic multi-body interactions and global structural constraints. Results: We propose the Relation-Aware Graph-Hypergraph Network (RA-GHN), a framework that unifies graph-based pairwise modeling with biophysical hypergraphs to capture high-order cooperative effects. Its heterogeneous hypergraph uses epitope-paratope interaction patches, spatial structural clusters, and high-degree hub structures. A dual-pathway architecture integrates a Relation-Aware Equivariant Graph Network with an Equivariant Hypergraph Encoder while preserving SE(3)-equivariance. We also introduce the Structure-Sequence Decoupling Index, which balances structural reliability with sequence novelty. Experiments show that RA-GHN outperforms state-of-the-art baselines across standard benchmarks and affinity optimization tasks.