A dual diffusion model enables 3D molecule generation and lead optimization based on target pockets.

Area of Science:

• Computational chemistry
• Drug discovery
• Artificial intelligence in medicine

Background:

• Structure-based generative chemistry is crucial for computer-aided drug discovery.
• Traditional methods face computational inefficiency, and current machine learning models have sampling bottlenecks.

Purpose of the Study:

• To develop an efficient conditional deep generative model for 3D molecule generation tailored to specific protein targets.
• To overcome the limitations of existing in silico drug design techniques.

Main Methods:

• Developed PMDM, a conditional equivariant diffusion model incorporating local and global molecular dynamics.
• Utilized protein information to guide the molecule generation process efficiently.
• Evaluated PMDM against baseline models using multiple metrics.

Main Results:

• PMDM demonstrated superior performance compared to existing models across various evaluation metrics.
• Applied PMDM to optimize lead compounds for SARS-CoV-2 main protease (Mpro) and Cyclin-dependent Kinase 2 (CDK2).
• Synthesized and tested optimized molecules, showing enhanced in-vitro activity against CDK2.

Conclusions:

• PMDM offers an efficient and effective approach for structure-based 3D molecule generation in drug discovery.
• The model shows promise for real-world applications, including lead compound optimization with improved target activity.