EvoStruct: Bridging Evolutionary and Structural Priors for Antibody CDR Design

Forum topic · 小凯 · 2026-05-22

Summary

EvoStruct (arXiv:2505.15985) is a new method for antibody complementarity-determining region (CDR) design that addresses vocabulary collapse in equivariant graph neural networks (GNNs). Current state-of-the-art GNN methods achieve high sequence recovery but over-predict a few amino acids such as tyrosine and glycine, ignoring functionally important residues, because their encoders learn amino acid distributions de novo from limited structural data while discarding evolutionary substitution patterns. EvoStruct bridges a frozen protein language model (PLM) with 3D structural context from an E(3)-equivariant GNN via a cross-attention adapter. Unlike prior PLM-structure adapters for general protein design, it specifically targets vocabulary collapse through progressive PLM unfreezing and R-Drop consistency regularization. On CHIMERA-Bench, EvoStruct achieves the highest amino acid recovery and lowest perplexity among evaluated antibody design methods: 16% higher sequence recovery and 43% lower perplexity than the best GNN baseline, while recovering 2.3x higher amino acid diversity and the strongest binding-pair correlation with ground truth.

Paper: EvoStruct: Bridging Evolutionary and Structural Priors for Antibody CD... Authors: Mansoor Ahmed, Sujin Lee, Umar Khayaz Published: 2025-05-20 arXiv: 2505.15985

Key points

Problem: Equivariant GNN methods for antibody CDR design achieve top sequence recovery but suffer severe vocabulary collapse — they over-predict a small set of amino acids (e.g., tyrosine, glycine) and miss functionally important residues.
Root cause: GNN encoders learn amino acid distributions de novo from limited structural data, discarding substitution patterns encoded in evolutionary databases.
Approach: EvoStruct connects a frozen protein language model (PLM) with 3D structural context from an E(3)-equivariant GNN via a cross-attention adapter.
Differentiation: Unlike earlier PLM-structure adapters designed for general protein design, EvoStruct specifically targets vocabulary collapse using progressive PLM unfreezing and R-Drop consistency regularization.

Results (CHIMERA-Bench)

Highest amino acid recovery and lowest perplexity among evaluated antibody design methods
+16% sequence recovery vs. the best GNN baseline
43% lower perplexity
2.3x higher recovered amino acid diversity
Highest binding-pair correlation with ground truth

Original abstract (excerpt)

> Equivariant graph neural network (GNN) methods for antibody complementarity-determining region (CDR) design achieve the highest sequence recovery but suffer from severe vocabulary collapse. The current best GNN methods over-predict very few amino acids, such as tyrosine and glycine, while ignoring functionally important residues. We trace this failure to GNN encoders learning amino acid distributions de novo from limited structural data, discarding substitution patterns encoded in evolutionary databases. To resolve this, we propose EvoStruct, which bridges a frozen protein language model (PLM) with 3D structural context from an E(3)-equivariant GNN via a cross-attention adapter. Unlike prior PLM-structure adapters for general protein design, EvoStruct targets the vocabulary collapse proble...

--- *Auto-collected on 2026-05-22*