Antibody-Antigen Docking and Design via Hierarchical Equivariant Refinement

Wengong Jin,Regina Barzilay,Tommi Jaakkola

DOI: https://doi.org/10.48550/arXiv.2207.06616

2022-07-14

Abstract:Computational antibody design seeks to automatically create an antibody that binds to an antigen. The binding affinity is governed by the 3D binding interface where antibody residues (paratope) closely interact with antigen residues (epitope). Thus, predicting 3D paratope-epitope complex (docking) is the key to finding the best paratope. In this paper, we propose a new model called Hierarchical Equivariant Refinement Network (HERN) for paratope docking and design. During docking, HERN employs a hierarchical message passing network to predict atomic forces and use them to refine a binding complex in an iterative, equivariant manner. During generation, its autoregressive decoder progressively docks generated paratopes and builds a geometric representation of the binding interface to guide the next residue choice. Our results show that HERN significantly outperforms prior state-of-the-art on paratope docking and design benchmarks.

Biomolecules,Machine Learning

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the key challenges in antibody - antigen docking and design. Specifically, the paper focuses on how to automatically create an antibody (i.e., the antigen - binding site) that can bind to a specific antigen region (i.e., the epitope). The binding affinity between the antibody and the antigen depends on the 3D binding interface, where the antibody residues (antigen - binding site) interact closely with the antigen residues (epitope). Therefore, predicting the 3D antigen - binding site - epitope complex (docking) is crucial for finding the optimal antigen - binding site.

The paper proposes a new model - the Hierarchical Equivariant Refinement Network (HERN) for antigen - binding site docking and design. During the docking process, HERN uses a hierarchical message - passing network to predict atomic forces and refines the binding complex in an iterative and equivariant manner. During the generation process, its autoregressive decoder gradually docks the generated antigen - binding site and constructs a geometric representation of the binding interface to guide the selection of the next residue. Experimental results show that HERN significantly outperforms previous state - of - the - art methods in antigen - binding site docking and design benchmark tests.