Abstract:We introduce AbDiffuser, an equivariant and physics-informed diffusion model for the joint generation of antibody 3D structures and sequences. AbDiffuser is built on top of a new representation of protein structure, relies on a novel architecture for aligned proteins, and utilizes strong diffusion priors to improve the denoising process. Our approach improves protein diffusion by taking advantage of domain knowledge and physics-based constraints; handles sequence-length changes; and reduces memory complexity by an order of magnitude, enabling backbone and side chain generation. We validate AbDiffuser in silico and in vitro. Numerical experiments showcase the ability of AbDiffuser to generate antibodies that closely track the sequence and structural properties of a reference set. Laboratory experiments confirm that all 16 HER2 antibodies discovered were expressed at high levels and that 57.1% of the selected designs were tight binders.

What problem does this paper attempt to address?

### What problem does this paper attempt to solve? This paper aims to solve the problem of efficient antibody (immunoglobulin) generation. The specific goals include: 1. **Generate the 3D structure and sequence of functional antibodies**: - The paper introduces an equivariant and physically - informed diffusion model named AbDiffuser for jointly generating the three - dimensional structure and sequence of antibodies. This model is based on new protein structure representation methods, new architectures for aligning proteins, and strong diffusion priors to improve the denoising process. 2. **Improve generation efficiency and quality**: - AbDiffuser improves the protein diffusion process by leveraging domain knowledge and physical constraints, handling sequence length variations, and reducing the memory complexity by an order of magnitude, enabling the generation of backbones and side chains. - The model has demonstrated in in - vitro and in - vivo validations the ability to generate antibodies that are highly consistent with the sequence and structural characteristics of the reference set. 3. **Address the challenges in protein generation**: - When generating protein structures, appropriate equivariance and physical constraints need to be met while balancing modeling complexity and fidelity. Most existing models rely on equivariant transformers or graph neural networks to satisfy SE(3) equivariance, usually representing part of the protein geometry in the angular space. - AbDiffuser introduces a new general - purpose SE(3) equivariant neural network - Aligned Protein Mixer (APMixer), which implicitly models the inter - residue relationships, is particularly good at handling sequence length variations, and significantly reduces the memory footprint, making it possible to generate complete antibody structures. 4. **Experimentally verify the functionality of the generated antibodies**: - The researchers verified the antibodies generated by AbDiffuser through numerical experiments and laboratory experiments. Laboratory experiments showed that all 16 HER2 antibodies were successfully expressed, with 57.1% of the designs having high - affinity binding capabilities, and some of the designs even outperformed the anti - cancer drug Trastuzumab in terms of affinity. In summary, the goal of this paper is to develop an efficient and high - quality antibody generation model that can generate new antibodies with predefined biochemical properties on a computer, thereby accelerating the drug design process.

AbDiffuser: Full-Atom Generation of in vitro Functioning Antibodies

Retrieval Augmented Diffusion Model for Structure-informed Antibody Design and Optimization

De novo antibody design with SE(3) diffusion

Guiding diffusion models for antibody sequence and structure co-design with developability properties

Improving Antibody Design with Force-Guided Sampling in Diffusion Models

BetterBodies: Reinforcement Learning guided Diffusion for Antibody Sequence Design

Combining transformer and 3DCNN models to achieve co-design of structures and sequences of antibodies in a diffusional manner

Atomically accurate de novo design of single-domain antibodies

An adaptive autoregressive diffusion approach to design active humanized antibody and nanobody

Inverse folding for antibody sequence design using deep learning

De novo design of protein structure and function with RFdiffusion

AbODE: Ab Initio Antibody Design using Conjoined ODEs

AntiBARTy Diffusion for Property Guided Antibody Design

AntiFold: Improved antibody structure-based design using inverse folding

IsAb: a computational protocol for antibody design

Toward enhancement of antibody thermostability and affinity by computational design in the absence of antigen

AI-based IsAb2.0 for antibody design

ABodyBuilder3: Improved and scalable antibody structure predictions

Antibody-SGM, a Score-Based Generative Model for Antibody Heavy-Chain Design

AbFold -- an AlphaFold Based Transfer Learning Model for Accurate Antibody Structure Prediction

Antigen-Specific Antibody Design and Optimization with Diffusion-Based Generative Models for Protein Structures