Deep Generation Model Guided by the Docking Score for Active Molecular Design.

Area of Science:

• Computational chemistry
• Medicinal chemistry
• Artificial intelligence in drug discovery

Background:

• Deep generation models offer advantages in designing novel drug compounds.
• Generating molecules with specific high activity remains a significant challenge in drug discovery.

Purpose of the Study:

• To develop a conditional molecular generation model (COMG) for designing drug compounds with both novelty and high target activity.
• To integrate docking scores and 3D pharmacophore matching into the molecular generation process.

Main Methods:

• Utilized a conditional variational autoencoder architecture constrained by pharmacophore matching scores.
• Employed Bayesian optimization with docking scores to enhance target relevance.
• Incorporated a scaffold memory unit to improve structural diversity and mitigate similarity issues.

Main Results:

• COMG demonstrated improved structural diversity in generated molecules.
• The model effectively increased the proportion of target-relevant, drug-active molecules.
• COMG proved to be a valuable tool for drug design and discovery.

Conclusions:

• The COMG model successfully addresses the challenge of generating novel, high-activity drug compounds.
• Integrating computational scoring and structural diversity mechanisms enhances molecular generation for drug discovery.
• COMG represents a significant advancement in AI-driven drug design.