Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Towards More Accurate Full-Atom Antibody Co-Design

Jiayang Wu, Xingyi Zhang, Xiangyu Dong, Kun Xie, Ziqi Liu, Wensheng Gan, Sibo Wang, Le Song

TL;DR

Igformer tackles the challenge of end-to-end antibody co-design by modeling antibody–antigen interfaces with a novel inter-graph refinement pathway that fuses Approximate Personalized Propagation and a global Simplified Graph Transformer under $E(3)$-equivariance. The framework enables simultaneous optimization of 1D CDR sequences and 3D full-atom structures, validated across tasks from single-CDR design to full antibody design and complex structure prediction, with substantial improvements over state-of-the-art baselines in both sequence recovery and docking accuracy. Ablation analyses confirm that each component—APP, SGFormer, TM, AA, and the dual EMP architecture—contributes significantly to performance. The empirical gains across RAbD and IgFold benchmarks suggest Igformer’s multi-scale, geometry-aware design strategy can accelerate computational antibody design for therapeutic development.

Abstract

Antibody co-design represents a critical frontier in drug development, where accurate prediction of both 1D sequence and 3D structure of complementarity-determining regions (CDRs) is essential for targeting specific epitopes. Despite recent advances in equivariant graph neural networks for antibody design, current approaches often fall short in capturing the intricate interactions that govern antibody-antigen recognition and binding specificity. In this work, we present Igformer, a novel end-to-end framework that addresses these limitations through innovative modeling of antibody-antigen binding interfaces. Our approach refines the inter-graph representation by integrating personalized propagation with global attention mechanisms, enabling comprehensive capture of the intricate interplay between local chemical interactions and global conformational dependencies that characterize effective antibody-antigen binding. Through extensive validation on epitope-binding CDR design and structure prediction tasks, Igformer demonstrates significant improvements over existing methods, suggesting that explicit modeling of multi-scale residue interactions can substantially advance computational antibody design for therapeutic applications.

Towards More Accurate Full-Atom Antibody Co-Design

TL;DR

Igformer tackles the challenge of end-to-end antibody co-design by modeling antibody–antigen interfaces with a novel inter-graph refinement pathway that fuses Approximate Personalized Propagation and a global Simplified Graph Transformer under -equivariance. The framework enables simultaneous optimization of 1D CDR sequences and 3D full-atom structures, validated across tasks from single-CDR design to full antibody design and complex structure prediction, with substantial improvements over state-of-the-art baselines in both sequence recovery and docking accuracy. Ablation analyses confirm that each component—APP, SGFormer, TM, AA, and the dual EMP architecture—contributes significantly to performance. The empirical gains across RAbD and IgFold benchmarks suggest Igformer’s multi-scale, geometry-aware design strategy can accelerate computational antibody design for therapeutic development.

Abstract

Antibody co-design represents a critical frontier in drug development, where accurate prediction of both 1D sequence and 3D structure of complementarity-determining regions (CDRs) is essential for targeting specific epitopes. Despite recent advances in equivariant graph neural networks for antibody design, current approaches often fall short in capturing the intricate interactions that govern antibody-antigen recognition and binding specificity. In this work, we present Igformer, a novel end-to-end framework that addresses these limitations through innovative modeling of antibody-antigen binding interfaces. Our approach refines the inter-graph representation by integrating personalized propagation with global attention mechanisms, enabling comprehensive capture of the intricate interplay between local chemical interactions and global conformational dependencies that characterize effective antibody-antigen binding. Through extensive validation on epitope-binding CDR design and structure prediction tasks, Igformer demonstrates significant improvements over existing methods, suggesting that explicit modeling of multi-scale residue interactions can substantially advance computational antibody design for therapeutic applications.

Paper Structure

This paper contains 43 sections, 10 theorems, 55 equations, 6 figures, 12 tables, 1 algorithm.

Table of Contents

  1. Main
  2. Results
  3. IgFormer
  4. Task 1: CDR-H3 Design
  5. Task 2: Multiple CDR Design
  6. Task 3: Full Antibody Design
  7. Task 4: Complex Structure Prediction
  8. Ablation Study
  9. Conclusion
  10. Related Work
  11. Problem Formulation
  12. Method
  13. Initialization
  14. Equivariant Message Passing
  15. Inter-Graph Refinement
  16. ...and 28 more sections

Key Result

Theorem C.1

For any transformation $T \in E(3)$, we have $\boldsymbol{H}_i^{(l+1)}, T(\boldsymbol{X}_i^{(l+1)}) = \textit{EMP}\left(\boldsymbol{H}_i^{(l)}, T(\boldsymbol{X}_i^{(l)})\right)$, where $T(X) := \boldsymbol{Q}\boldsymbol{X} + \boldsymbol{b}$ denotes the E(3) transformation of $\boldsymbol{X}$.

Figures (6)

  • Figure 1: End-to-end antibody co-design task.
  • Figure 2: Framework of Igformer.
  • Figure 3: Antibody structures by Igformer & dyMEAN.
  • Figure 4: Illustration of antibody-antigen complex.
  • Figure 5: DockQ scores for varying $w$ values across four tasks, with the peak observed at $w=0.2$, emphasizing the importance of geometric distance in similairty matrix computation.
  • ...and 1 more figures

Theorems & Definitions (20)

  • Definition B.1
  • Theorem C.1
  • Remark C.2
  • Theorem C.3
  • Definition D.1: E(3)-equivariance and SE(3)-invariance
  • Lemma D.2: Linear Transformation of Coordinate Differences
  • proof
  • Lemma D.3: Geometry-Based Operations
  • proof
  • Theorem D.4: E(3)-Equivariance of Coordinates in EMP
  • ...and 10 more