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

The design of antibody complementarity-determining regions (CDRs) is crucial for therapeutic development, but existing methods struggle to capture the complexity of antibody-antigen interfaces due to their reliance on pairwise relationships. To address this limitation, researchers propose the Relation-Aware Graph-Hypergraph Network (RA-GHN), a framework that integrates graph-based pairwise modeling with biophysical hypergraphs to account for high-order cooperative effects.

The RA-GHN utilizes a heterogeneous hypergraph composed of epitope-paratope interaction patches, spatial structural clusters, and high-degree hub structures. A dual-pathway architecture within the RA-GHN combines a Relation-Aware Equivariant Graph Network with an Equivariant Hypergraph Encoder, both of which maintain SE(3)-equivariance.

To ensure balanced structural reliability and sequence novelty, the researchers introduce the Structure-Sequence Decoupling Index. The RA-GHN demonstrates superior performance over existing state-of-the-art baselines across standard benchmarks and affinity optimization tasks.