MolSnapper: Conditioning Diffusion for Structure-Based Drug Design.

Area of Science:

• Computational chemistry
• Medicinal chemistry
• Drug discovery

Background:

• Generative models show promise for molecular design but struggle with target binding.
• Controlling molecular design and incorporating prior knowledge are crucial for drug development.
• Tailoring molecules to specific binding sites remains a significant challenge in structure-based drug design.

Purpose of the Study:

• Introduce MolSnapper, a novel tool for structure-based drug design.
• Enable conditioning of diffusion models using expert knowledge (3D pharmacophores).
• Improve the generation of molecules with high binding affinity and specificity.

Main Methods:

• Developed MolSnapper to integrate 3D pharmacophores into diffusion models.
• Conditioned generative models using expert-defined pharmacophore constraints.
• Validated the method on the CrossDocked and Binding MOAD datasets.

Main Results:

• MolSnapper generates molecules better tailored to specific binding sites.
• Achieved high structural and chemical similarity to original molecules.
• Produced approximately twice as many valid molecules compared to alternative methods.

Conclusions:

• MolSnapper effectively integrates expert knowledge into structure-based drug design.
• The tool enhances the generation of high-quality drug candidates.
• MolSnapper offers a significant improvement in generating valid molecules for drug discovery.