Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Fast and Accurate Antibody Sequence Design via Structure Retrieval

Xingyi Zhang, Kun Xie, Ningqiao Huang, Wei Liu, Peilin Zhao, Sibo Wang, Kangfei Zhao, Biaobin Jiang

TL;DR

Antibody design often suffers from hallucinations when inferring CDR sequences from backbone structures via inverse folding. IgSeek addresses this by retrieving structurally similar CDR templates from a natural antibody database using a MEGNN-encoded vector space to guide sequence design, followed by ensemble-based sequence prediction from the retrieved templates. The approach yields faster structure retrieval than state-of-the-art methods and higher sequence-recovery accuracy across CDR types, with strong generalization to T-cell receptors and substantial efficiency gains (about $20$-fold faster inference on OAS-H3). This retrieval-augmented framework offers a scalable, practical pathway for rapid and accurate therapeutic protein design, including antibodies and receptors.

Abstract

Recent advancements in protein design have leveraged diffusion models to generate structural scaffolds, followed by a process known as protein inverse folding, which involves sequence inference on these scaffolds. However, these methodologies face significant challenges when applied to hyper-variable structures such as antibody Complementarity-Determining Regions (CDRs), where sequence inference frequently results in non-functional sequences due to hallucinations. Distinguished from prevailing protein inverse folding approaches, this paper introduces Igseek, a novel structure-retrieval framework that infers CDR sequences by retrieving similar structures from a natural antibody database. Specifically, Igseek employs a simple yet effective multi-channel equivariant graph neural network to generate high-quality geometric representations of CDR backbone structures. Subsequently, it aligns sequences of structurally similar CDRs and utilizes structurally conserved sequence motifs to enhance inference accuracy. Our experiments demonstrate that Igseek not only proves to be highly efficient in structural retrieval but also outperforms state-of-the-art approaches in sequence recovery for both antibodies and T-Cell Receptors, offering a new retrieval-based perspective for therapeutic protein design.

Fast and Accurate Antibody Sequence Design via Structure Retrieval

TL;DR

Antibody design often suffers from hallucinations when inferring CDR sequences from backbone structures via inverse folding. IgSeek addresses this by retrieving structurally similar CDR templates from a natural antibody database using a MEGNN-encoded vector space to guide sequence design, followed by ensemble-based sequence prediction from the retrieved templates. The approach yields faster structure retrieval than state-of-the-art methods and higher sequence-recovery accuracy across CDR types, with strong generalization to T-cell receptors and substantial efficiency gains (about -fold faster inference on OAS-H3). This retrieval-augmented framework offers a scalable, practical pathway for rapid and accurate therapeutic protein design, including antibodies and receptors.

Abstract

Recent advancements in protein design have leveraged diffusion models to generate structural scaffolds, followed by a process known as protein inverse folding, which involves sequence inference on these scaffolds. However, these methodologies face significant challenges when applied to hyper-variable structures such as antibody Complementarity-Determining Regions (CDRs), where sequence inference frequently results in non-functional sequences due to hallucinations. Distinguished from prevailing protein inverse folding approaches, this paper introduces Igseek, a novel structure-retrieval framework that infers CDR sequences by retrieving similar structures from a natural antibody database. Specifically, Igseek employs a simple yet effective multi-channel equivariant graph neural network to generate high-quality geometric representations of CDR backbone structures. Subsequently, it aligns sequences of structurally similar CDRs and utilizes structurally conserved sequence motifs to enhance inference accuracy. Our experiments demonstrate that Igseek not only proves to be highly efficient in structural retrieval but also outperforms state-of-the-art approaches in sequence recovery for both antibodies and T-Cell Receptors, offering a new retrieval-based perspective for therapeutic protein design.

Paper Structure

This paper contains 17 sections, 1 theorem, 14 equations, 8 figures, 5 tables, 3 algorithms.

Table of Contents

  1. Main
  2. Results
  3. IgSeek Approach
  4. IgSeek for CDR Structure Retrieval
  5. IgSeek for CDR Sequence Design
  6. Case Study
  7. Conclusion
  8. Related Work
  9. Preliminaries and Problem Formulation
  10. Methodology: IgSeek
  11. Multi Channel Equivariant Encoder
  12. Learning Objective and Sequence Generation
  13. Analysis
  14. Datasets and Labels
  15. Implementation Details
  16. ...and 2 more sections

Key Result

Theorem 1

For any transformation $g \in E(3)$, we have $\boldsymbol{h}_i, T_{\mathcal{Y}}(g) \boldsymbol{X}_i^{(L)} = \text{MEGNN} \left( \boldsymbol{h}_i^{(0)}, T_{\mathcal{X}}(g) \boldsymbol{X}_i^{(0)}, G \right)$, where $T_{\mathcal{X}}$ and $T_{\mathcal{Y}}$$:= \boldsymbol{R} \boldsymbol{X} + \boldsymbol{

Figures (8)

  • Figure 1: The Framework IgSeek: (a) Pre-train an MEGNN encoder by a self-supervised learning task. (b) Construct a CDR vector database. (c) Sequence generation by K-NN search.
  • Figure 2: The comparison of average AAR and inference speed. (a) AAR in SAbDab-2024 dataset. (b) AAR in STCRDab. (c) Inference speed.
  • Figure 3: IgSeek vs. FoldSeek in CDR retrieval.
  • Figure 4: Embeddings of CDRs in the SAbDab-before-2024 datasets projected onto 2D Space.
  • Figure 5: A Case study using 8W8R CDR-L1 as an example.
  • ...and 3 more figures

Theorems & Definitions (2)

  • Theorem 1
  • proof