Deep Lead Optimization: Leveraging Generative AI for Structural Modification.

Area of Science:

• Computational Chemistry
• Medicinal Chemistry
• Artificial Intelligence

Background:

• Deep learning models accelerate drug discovery by generating novel molecules.
• Lead optimization refines existing molecules into drug candidates.
• Classifying optimization methods is crucial for understanding their applications.

Purpose of the Study:

• To systematically review and categorize deep learning-based lead optimization methods.
• To focus on structure-directed optimization techniques and their specific tasks.
• To provide practical guidance for utilizing generative AI in drug design.

Main Methods:

• Categorization of lead optimization into goal-directed and structure-directed approaches.
• Systematic review of computational methods for structure-directed optimization.
• Analysis of tasks: fragment replacement, linker design, scaffold hopping, side-chain decoration.

Main Results:

• Identified four key tasks within structure-directed lead optimization.
• Discussed motivations, data construction, and current developments for each task.
• Classified both goal-directed and structure-directed methods using optimization taxonomy.

Conclusions:

• Structure-directed optimization is vital for practical drug discovery but less explored.
• Generative AI tools offer significant potential for molecular structure modification.
• A reference protocol is proposed to bridge the gap between AI advancements and experimental chemistry.