Joint Design of Protein Surface and Structure Using a Diffusion Bridge Model

TL;DR

PepBridge addresses the challenge of jointly designing protein surfaces and structures for receptor binding by introducing a diffusion bridge that maps receptor surface distributions to complementary ligand surfaces, followed by SE(3) backbone diffusion and a Shape-Frame Matching Network to ensure geometric and biochemical coherence. The framework integrates surface-conditioned surface generation, multi-modal bottom-structure diffusion, and a bidirectional surface-frame interaction scheme within a unified training loss. Empirical results on PepMerge demonstrate improved diversity, affinity, stability, and RMSD over backbone-only and full-atom baselines, validating the approach's effectiveness for top-down protein design. This work advances practical protein engineering by enabling coherent, receptor-aware design of both surface geometry and underlying backbone architecture.

Abstract

Protein-protein interactions (PPIs) are governed by surface complementarity and hydrophobic interactions at protein interfaces. However, designing diverse and physically realistic protein structure and surfaces that precisely complement target receptors remains a significant challenge in computational protein design. In this work, we introduce PepBridge, a novel framework for the joint design of protein surface and structure that seamlessly integrates receptor surface geometry and biochemical properties. Starting with a receptor surface represented as a 3D point cloud, PepBridge generates complete protein structures through a multi-step process. First, it employs denoising diffusion bridge models (DDBMs) to map receptor surfaces to ligand surfaces. Next, a multi-model diffusion model predicts the corresponding structure, while Shape-Frame Matching Networks ensure alignment between surface geometry and backbone architecture. This integrated approach facilitates surface complementarity, conformational stability, and chemical feasibility. Extensive validation across diverse protein design scenarios demonstrates PepBridge's efficacy in generating structurally viable proteins, representing a significant advancement in the joint design of top-down protein structure.

Figures (4)

• Figure 1: Top-down view of the receptor-peptide complex.
• Figure 2: Illustration of the PepBridge architecture for joint surface-structure peptide generation. (a) The model processes receptor-ligand pairs through a top-down structure comprising molecular surface and frame components. (b) Two specialized diffusion models are employed simultaneously. A diffusion bridge model leverages the receptor surface as the starting point to generate peptide surfaces. An SE(3) diffusion model shoulders the responsibility of frame construction, which incorporates translation and torsion angles. (c) A surface-frame matching network facilitates the interaction between creased structures, while multi-modal diffusion reconstructs the complete peptide structure.
• Figure 3: Visualization of generated peptides by our PepBridge. Top: Generated peptides (in orange) for receptors (in purple). The PDB ID of the receptors are 5DJA, 1UJ0, 2R02, and 3AVC. Bottom: The generated backbone structure and surface. The ground-truth surface structure (in black) and generated surface (in orange) are shown to compare the ability of the interface caption.
• Figure 4: Visualization of generated peptides by PepBridge. Top: Generated peptides (in orange) for receptors (in purple). The PDB ID of the receptors are 4CC2, 3AV9, 1B07, and 2XS1. Bottom: The generated backbone structure and surface. The ground-truth surface structure (in black) and generated surface (in orange) are shown to compare the ability of interface caption.