Integrating structure-based approaches in generative molecular design.

Area of Science:

• Computational chemistry
• Drug discovery
• Molecular modeling

Background:

• Generative molecular design accelerates drug discovery by exploring larger chemical spaces than traditional virtual screening.
• Current generative models primarily use small-molecule data, limiting their potential.
• Incorporating protein structure into de novo molecule generation is a promising advancement.

Purpose of the Study:

• To review recent approaches that integrate protein structure into de novo molecule optimization.
• To categorize these methods based on structure integration principles (distribution learning vs. goal-directed optimization) and model explicitness (explicit vs. implicit).
• To provide insights into future research directions in structure-based generative molecular design.

Main Methods:

• Summarizing recent generative molecular design approaches that incorporate protein structural information.
• Categorizing methods based on whether they use distribution learning or goal-directed optimization.
• Analyzing whether protein structure integration is explicit or implicit within the generative model.

Main Results:

• Identified two main principles for integrating protein structure: distribution learning and goal-directed optimization.
• Differentiated approaches based on protein structure explicitness (explicit vs. implicit).
• Highlighted the potential of structure-informed generative models to enhance on-target binding affinity predictions.

Conclusions:

• Structure-based generative models offer a powerful paradigm for drug discovery.
• Future work should focus on refining explicit and implicit structure integration methods.
• Optimizing generative models with protein structure data can significantly improve drug design efficiency and success rates.