Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Binding-Adaptive Diffusion Models for Structure-Based Drug Design

Zhilin Huang, Ling Yang, Zaixi Zhang, Xiangxin Zhou, Yu Bao, Xiawu Zheng, Yuwei Yang, Yu Wang, Wenming Yang

TL;DR

BindDM introduces a binding-adaptive diffusion framework that at each denoising step extracts a binding-relevant subcomplex from the protein-ligand complex and processes it with SE(3)-equivariant networks, coupled via cross-hierarchy interaction nodes to fuse context for target-aware 3D ligand generation. On CrossDocked2020, BindDM delivers more realistic 3D structures and stronger binding affinities than competitive baselines, achieving notable gains in Avg. Vina Score and high-affinity ligands while maintaining drug-like properties. The approach emphasizes binding-specific substructure mining and iterative context exchange, offering a single-stage diffusion alternative that leverages binding clues for pocket-specific design in SBDD. This work advances practical drug design by producing candidates with enhanced target engagement and plausible pharmacological properties.

Abstract

Structure-based drug design (SBDD) aims to generate 3D ligand molecules that bind to specific protein targets. Existing 3D deep generative models including diffusion models have shown great promise for SBDD. However, it is complex to capture the essential protein-ligand interactions exactly in 3D space for molecular generation. To address this problem, we propose a novel framework, namely Binding-Adaptive Diffusion Models (BindDM). In BindDM, we adaptively extract subcomplex, the essential part of binding sites responsible for protein-ligand interactions. Then the selected protein-ligand subcomplex is processed with SE(3)-equivariant neural networks, and transmitted back to each atom of the complex for augmenting the target-aware 3D molecule diffusion generation with binding interaction information. We iterate this hierarchical complex-subcomplex process with cross-hierarchy interaction node for adequately fusing global binding context between the complex and its corresponding subcomplex. Empirical studies on the CrossDocked2020 dataset show BindDM can generate molecules with more realistic 3D structures and higher binding affinities towards the protein targets, with up to -5.92 Avg. Vina Score, while maintaining proper molecular properties. Our code is available at https://github.com/YangLing0818/BindDM

Binding-Adaptive Diffusion Models for Structure-Based Drug Design

TL;DR

BindDM introduces a binding-adaptive diffusion framework that at each denoising step extracts a binding-relevant subcomplex from the protein-ligand complex and processes it with SE(3)-equivariant networks, coupled via cross-hierarchy interaction nodes to fuse context for target-aware 3D ligand generation. On CrossDocked2020, BindDM delivers more realistic 3D structures and stronger binding affinities than competitive baselines, achieving notable gains in Avg. Vina Score and high-affinity ligands while maintaining drug-like properties. The approach emphasizes binding-specific substructure mining and iterative context exchange, offering a single-stage diffusion alternative that leverages binding clues for pocket-specific design in SBDD. This work advances practical drug design by producing candidates with enhanced target engagement and plausible pharmacological properties.

Abstract

Structure-based drug design (SBDD) aims to generate 3D ligand molecules that bind to specific protein targets. Existing 3D deep generative models including diffusion models have shown great promise for SBDD. However, it is complex to capture the essential protein-ligand interactions exactly in 3D space for molecular generation. To address this problem, we propose a novel framework, namely Binding-Adaptive Diffusion Models (BindDM). In BindDM, we adaptively extract subcomplex, the essential part of binding sites responsible for protein-ligand interactions. Then the selected protein-ligand subcomplex is processed with SE(3)-equivariant neural networks, and transmitted back to each atom of the complex for augmenting the target-aware 3D molecule diffusion generation with binding interaction information. We iterate this hierarchical complex-subcomplex process with cross-hierarchy interaction node for adequately fusing global binding context between the complex and its corresponding subcomplex. Empirical studies on the CrossDocked2020 dataset show BindDM can generate molecules with more realistic 3D structures and higher binding affinities towards the protein targets, with up to -5.92 Avg. Vina Score, while maintaining proper molecular properties. Our code is available at https://github.com/YangLing0818/BindDM
Paper Structure (33 sections, 1 theorem, 12 equations, 6 figures, 7 tables, 2 algorithms)

This paper contains 33 sections, 1 theorem, 12 equations, 6 figures, 7 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Related Work
  3. Structure-Based Drug Design
  4. Diffusion Models for SBDD
  5. Preliminary
  6. The Proposed BindDM
  7. Binding-Adaptive Subcomplex Extraction
  8. 3D Equivariant Complex-Subcomplex Processing
  9. Iterative Cross-Hierarchy Interaction
  10. Training and Sampling
  11. Experiments
  12. Experimental Settings
  13. Datasets and Baseline Methods
  14. Evaluation
  15. Main Results
  16. ...and 18 more sections

Key Result

Proposition 1

Denoting SE(3)-transformation as $T$, we can achieve invariant likelihood w.r.t$T$ on both the protein-ligand complex and its subcomplex: $p_\theta(T{\mathbf{M}}_0|T{\mathbf{P}}) = p_\theta({\mathbf{M}}_0|{\mathbf{P}})$ if we shift the Center of Mass (CoM) of protein atoms to zero and parameterize t

Figures (6)

  • Figure 1: BindDM extracts subcomplex from protein-ligand complex, and utilizes it to enhance the binding-adaptive 3D molecule generation in complex.
  • Figure 2: The overview of BindDM.
  • Figure 3: The overview of the gated transmission module.
  • Figure 4: The generated ligand molecules of TargetDiff guan2023target and BindDM for the given protein pockets. Carbon atoms in ligands generated by TargetDiff and BindDM are visualized in green and orange, respectively. We report Vina Score, QED, SA for each molecule.
  • Figure 5: The subcomplex prediction accuracy of BASE in each layer of the de-nosing network.
  • ...and 1 more figures

Theorems & Definitions (1)

  • Proposition 1