Relation-Aware Equivariant Graph Networks for Epitope-Unknown Antibody Design and Specificity Optimization

TL;DR

This work tackles the problem of designing antigen-specific antibody CDRs when epitopes are unknown and specificity must be preserved. It introduces RAAD, a one-shot framework based on an Attributed Heterogeneous Graph (AHG) and Relation-Aware Equivariant Graph Networks (RA-EGN) to jointly model sequence and structure while dynamically updating antigen–antibody interactions through multiple edge relations. A new SP-score is proposed to quantify specificity, augmented by a contrastive mutual-information loss and Iterative Target Augmentation to steer optimization toward target selectivity without sacrificing realism. Across extensive experiments on SAbDab and RAbD benchmarks, RAAD outperforms state-of-the-art baselines in CDR sequence/structure modeling, antigen-binding CDR-H3 generation, and specificity optimization, demonstrating strong robustness to CDR length and input context and highlighting the practical potential for epitope-unknown antibody design.

Abstract

Antibodies are Y-shaped proteins that protect the host by binding to specific antigens, and their binding is mainly determined by the Complementary Determining Regions (CDRs) in the antibody. Despite the great progress made in CDR design, existing computational methods still encounter several challenges: 1) poor capability of modeling complex CDRs with long sequences due to insufficient contextual information; 2) conditioned on pre-given antigenic epitopes and their static interaction with the target antibody; 3) neglect of specificity during antibody optimization leads to non-specific antibodies. In this paper, we take into account a variety of node features, edge features, and edge relations to include more contextual and geometric information. We propose a novel Relation-Aware Antibody Design (RAAD) framework, which dynamically models antigen-antibody interactions for co-designing the sequences and structures of antigen-specific CDRs. Furthermore, we propose a new evaluation metric to better measure antibody specificity and develop a contrasting specificity-enhancing constraint to optimize the specificity of antibodies. Extensive experiments have demonstrated the superior capability of RAAD in terms of antibody modeling, generation, and optimization across different CDR types, sequence lengths, pre-training strategies, and input contexts.

Figures (11)

• Figure 1: Correlation of the performance of sequence generation (Left) and structure generation (Right) w.r.t amino acid position in CDR-H3 for different lengths $L$ from SAbDab.
• Figure 2: Structure of an antigen-antibody complex, where the antibody is symmetrically Y-shape, and we mainly focus on three CDRs of the variable domain in the heavy chain.
• Figure 3: A high-level overview of our RAAD framework. The antigen-antibody complex is constructed as an attributed heterogeneous graph, which is passed through multilayer Relation-Aware Equivariant Graph Networks (RA-EGN). The outputs are fed to sequence generation and structure predictor to generate both the sequence and structure of CDRs in the antibody.
• Figure 4: Each RA-EGN layer aggregates messages and jointly updates features, coordinates, and relations.
• Figure 5: A comparison between RAAD and previous methods in antigen-antibody interaction modeling. Their main differences include whether the interactions are dynamically updatable and whether antigenic epitopes are pre-given or adaptively learned.

Theorems & Definitions (2)

• Theorem 1
• Theorem 1