SGEDiff: a subgraph-enriched diffusion model for structure-based 3D molecular generation.

Area of Science:

• Computational chemistry
• Drug discovery
• Molecular modeling

Background:

• Structure-based molecular generation is key in computer-aided drug discovery.
• Current diffusion models struggle with protein-ligand representation and predefined binding pockets.

Purpose of the Study:

• To introduce SGEDiff, a subgraph-enriched generative framework for 3D molecule generation.
• To address limitations in existing diffusion-based models for drug design.

Main Methods:

• SGEDiff hierarchically fuses subgraph and global graph representations.
• An integrated pocket prediction module identifies binding regions without predefined coordinates.
• The model captures local binding patterns and key structural features of protein pockets.

Main Results:

• SGEDiff outperforms baseline diffusion models in generating high-affinity molecules.
• The framework demonstrates improved success rates in de novo drug design for novel protein targets.
• Experimental results validate the model's effectiveness across diverse targets.

Conclusions:

• SGEDiff advances structure-based drug discovery by enabling effective de novo design.
• The model's ability to predict binding pockets broadens its applicability to new protein targets.
• SGEDiff offers a promising approach for generating novel drug candidates.