3D molecular generative framework for interaction-guided drug design.

Area of Science:

• Computational chemistry
• Drug discovery
• Artificial intelligence in medicine

Background:

• Deep generative models show promise for accelerating drug design.
• Existing models struggle with generalization and data limitations, hindering innovative drug designs and leading to unfavorable interactions with unseen target proteins.

Purpose of the Study:

• To propose an interaction-aware 3D molecular generative framework for interaction-guided drug design within target binding pockets.
• To improve generalizability and innovation in drug design, especially with limited experimental data.

Main Methods:

• Developed an interaction-aware 3D molecular generative framework.
• Leveraged universal patterns of protein-ligand interactions as prior knowledge.
• Assessed model performance on generated ligands for unseen targets, evaluating binding pose stability, affinity, geometric patterns, diversity, and novelty.

Main Results:

• The proposed framework demonstrates high generalizability even with limited experimental data.
• Generated ligands exhibit favorable interactions and desirable properties for unseen target proteins.
• The model successfully designed potential mutant-selective inhibitors, showcasing its applicability in structure-based drug design.

Conclusions:

• The interaction-aware generative framework effectively addresses limitations of existing models in drug design.
• The approach enables interaction-guided drug design, improving innovation and generalizability.
• This method holds significant potential for advancing structure-based drug design and accelerating the discovery of novel therapeutics.