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

Published on: January 15, 2024

Impact Score: 7.6

Arxiv code: Arxiv:2402.18583

Summary

• What is new: Introduces Binding-Adaptive Diffusion Models (BindDM) for capturing protein-ligand interactions in 3D molecule generation.
• Why this is important: Existing 3D deep generative models for structure-based drug design struggle to accurately capture essential protein-ligand interactions.
• What the research proposes: BindDM extracts essential binding sites (subcomplex) and uses SE(3)-equivariant neural networks to infuse binding interaction information into the molecule generation process.
• Results: BindDM generates molecules with more realistic 3D structures and higher binding affinities, achieving up to -5.92 Avg. Vina Score.

Technical Details

Technological frameworks used: BindDM

Models used: SE(3)-equivariant neural networks

Data used: CrossDocked2020 dataset

Potential Impact

Pharmaceutical companies, biotech startups in drug discovery, companies in molecular simulation software.

Want to implement this idea in a business?

We have generated a startup concept here: BindOptics.