PPDiff: Diffusing in Hybrid Sequence-Structure Space for Protein-Protein Complex Design

2025-06-13

AI Summary

To address the challenge of *de novo* designing high-affinity binding proteins for arbitrary protein targets, this paper introduces PPDiff, a non-autoregressive diffusion model that jointly generates binder sequences and 3D conformations in sequence–structure latent space. We propose SSINC, a novel sequence–structure interleaved architecture integrating causal attention, k-nearest-neighbor dynamic graph networks, and global self-attention. We construct PPBench—the first large-scale, general-purpose protein–protein complex benchmark—comprising 706,000 complexes. PPDiff adopts a multi-stage pretraining–fine-tuning paradigm, combining diffusion-based generative modeling with equivariant representation learning. On pretraining reconstruction tasks and downstream applications—including mini-binder design and antigen–antibody co-design—PPDiff achieves success rates of 50.00%, 23.16%, and 16.89%, respectively, substantially outperforming existing methods.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Large Multimodal Models (LMMs)
• Humans and AI: Human-Aware Planning and Behavior Prediction

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Designing protein-binding proteins with high affinity is critical in biomedical research and biotechnology. Despite recent advancements targeting specific proteins, the ability to create high-affinity binders for arbitrary protein targets on demand, without extensive rounds of wet-lab testing, remains a significant challenge. Here, we introduce PPDiff, a diffusion model to jointly design the sequence and structure of binders for arbitrary protein targets in a non-autoregressive manner. PPDiffbuilds upon our developed Sequence Structure Interleaving Network with Causal attention layers (SSINC), which integrates interleaved self-attention layers to capture global amino acid correlations, k-nearest neighbor (kNN) equivariant graph layers to model local interactions in three-dimensional (3D) space, and causal attention layers to simplify the intricate interdependencies within the protein sequence. To assess PPDiff, we curate PPBench, a general protein-protein complex dataset comprising 706,360 complexes from the Protein Data Bank (PDB). The model is pretrained on PPBenchand finetuned on two real-world applications: target-protein mini-binder complex design and antigen-antibody complex design. PPDiffconsistently surpasses baseline methods, achieving success rates of 50.00%, 23.16%, and 16.89% for the pretraining task and the two downstream applications, respectively.

Problem

• Design high-affinity protein binders for arbitrary targets
• Jointly model protein sequence and structure non-autoregressively
• Overcome wet-lab dependency for protein-protein complex design

Innovation

• Diffusion model for sequence-structure joint design
• SSINC network with interleaved attention layers
• kNN graph layers for 3D local interactions