A Scaffold-based Deep Generative Model Considering Molecular Stereochemical Information.

Area of Science:

• Medicinal Chemistry
• Computational Chemistry
• Drug Discovery

Background:

• Scaffold-based molecular design is crucial for lead compound discovery.
• Existing deep learning models often neglect 3D structure and stereochemistry, limiting scaffold hopping.
• Natural products present unique stereochemical challenges for molecular generation.

Purpose of the Study:

• To optimize a scaffold-based molecular generation model for improved scaffold hopping.
• To incorporate 3D conformational and pharmacophore matching for scaffold generalization.
• To enable the generation of chiral molecules, particularly relevant for natural products.

Main Methods:

• Optimized a scaffold-based molecular generation model using deep learning.
• Integrated Real-time Ultrafast Shape Recognition with Pharmacophore constraints (USRCAT) for scaffold searching.
• Utilized identified similar scaffolds for scaffold hopping and new molecule generation.

Main Results:

• The optimized model successfully performed scaffold hopping by identifying 3D similar scaffolds.
• Generated novel molecules retained the same chirality as the input scaffolds.
• Analysis confirmed the model's capability in generating diverse molecular structures and physicochemical properties.

Conclusions:

• The enhanced model facilitates exploration of wider chemical spaces for lead optimization.
• It provides a valuable tool for screening virtual compound libraries with preserved stereochemistry.
• This approach advances scaffold-based drug design by considering 3D conformation and chirality.