Target-aware 3D molecular generation based on guided equivariant diffusion.

Area of Science:

• Computational chemistry
• Molecular modeling
• Drug discovery

Background:

• Current molecular generation models for structure-based drug design (SBDD) often yield unrealistic 3D molecules.
• Existing methods struggle with structural feasibility and incorporating essential drug-like properties during generation.

Purpose of the Study:

• To introduce DiffGui, an E(3)-equivariant diffusion model for SBDD.
• To address the limitations of existing models by integrating bond diffusion and property guidance.

Main Methods:

• DiffGui employs a combination of atom diffusion and bond diffusion for concurrent generation of molecular structures.
• The model incorporates property guidance to integrate binding affinity and drug-like properties into training and sampling.
• Utilizes target-conditioning for structure-based molecular generation.

Main Results:

• DiffGui outperforms existing methods in generating molecules with high binding affinity and rational chemical structures.
• The model successfully generates molecules with desirable drug-like properties.
• Ablation studies validate the significant contributions of the bond diffusion and property guidance modules.

Conclusions:

• DiffGui effectively generates realistic and drug-like 3D molecules for SBDD.
• The model shows promise for both de novo drug design and lead optimization.
• Wet-lab validation confirms the practical utility of DiffGui in drug discovery pipelines.