[Paper Review] Conditional Antibody Design as 3D Equivariant Graph Translation

Xiangzhe Kong, Wenbing Huang+1 more|arXiv (Cornell University)|Aug 12, 2022

Monoclonal and Polyclonal Antibodies Research25 citations

TL;DR

The paper introduces MEAN, an end-to-end framework that co-designs 1D CDR sequences and 3D structures in an antibody–antigen context using E(3)-equivariant graph translation, achieving state-of-the-art results.

ABSTRACT

Antibody design is valuable for therapeutic usage and biological research. Existing deep-learning-based methods encounter several key issues: 1) incomplete context for Complementarity-Determining Regions (CDRs) generation; 2) incapability of capturing the entire 3D geometry of the input structure; 3) inefficient prediction of the CDR sequences in an autoregressive manner. In this paper, we propose Multi-channel Equivariant Attention Network (MEAN) to co-design 1D sequences and 3D structures of CDRs. To be specific, MEAN formulates antibody design as a conditional graph translation problem by importing extra components including the target antigen and the light chain of the antibody. Then, MEAN resorts to E(3)-equivariant message passing along with a proposed attention mechanism to better capture the geometrical correlation between different components. Finally, it outputs both the 1D sequences and 3D structure via a multi-round progressive full-shot scheme, which enjoys more efficiency and precision against previous autoregressive approaches. Our method significantly surpasses state-of-the-art models in sequence and structure modeling, antigen-binding CDR design, and binding affinity optimization. Specifically, the relative improvement to baselines is about 23% in antigen-binding CDR design and 34% for affinity optimization.

Motivation & Objective

• Motivate antibody design as a conditional 3D translation problem that includes antigen and light chain context.
• Develop an end-to-end E(3)-equivariant model to generate CDR sequences and 3D conformations.
• Enable efficient, non-autoregressive (full-shot) decoding to improve inference speed and accuracy.
• Demonstrate improvements in sequence/structure modeling, antigen-binding CDR design, and affinity optimization.

Proposed method

• Formulate antibody design as conditional graph translation over an antibody–antigen complex graph.
• Propose MEAN, an E(3)-equivariant multi-channel graph neural network with internal context encoder and external attentive encoder.
• Use a progressive full-shot decoding scheme to predict CDR sequences and coordinates in multiple rounds.
• Incorporate global nodes and inter-component edges to capture intra- and inter-molecular interactions.
• Train with combined sequence and structure losses, leveraging an end-to-end equivariant framework.

Experimental results

Research questions

• RQ1 How can antibody design be cast as a 3D equivariant graph translation problem given antigen and light chain context?
• RQ2 Does MEAN improve both 1D CDR recovery and 3D structure accuracy compared to state-of-the-art baselines?
• RQ3 Can progressive full-shot decoding provide efficient, accurate, non-autoregressive CDR design and affinity optimization?

Key findings

• MEAN substantially outperforms baselines in both 1D sequence recovery (AAR) and 3D structure RMSD across CDR-H1, CDR-H2, and CDR-H3.
• On CDR-H1/H2/H3 tasks, MEAN achieves higher AAR and lower RMSD than RefineGNN and variants that exclude antigen/light chain context.
• In antigen-binding CDR-H3 design, MEAN attains higher AAR and TM-scores approaching 1 (TM-score near 0.98–0.99) with near-native structures.
• MEAN demonstrates notable affinity-optimization gains, achieving lower predicted ΔΔG values compared with baselines.
• Progressive full-shot decoding offers 2–5x speedups over iterative refinement while maintaining or improving accuracy.