CMOMO: a deep multi-objective optimization framework for constrained molecular multi-property optimization.

Area of Science:

• Computational chemistry
• Artificial intelligence in drug discovery
• Molecular modeling

Background:

• Molecular optimization is crucial for drug development but challenging due to multi-property optimization and strict drug-like constraints.
• Existing artificial intelligence methods often overlook these constraints, limiting the quality of optimized molecules.
• There is a need for advanced frameworks that can handle both property objectives and constraint compliance.

Purpose of the Study:

• To propose a deep multi-objective optimization framework, CMOMO, for constrained molecular multi-property optimization.
• To address the limitations of existing methods by incorporating a dynamic constraint handling strategy.
• To enhance the generation of high-quality molecules that satisfy both property goals and drug-like criteria.

Main Methods:

• Developed a two-stage deep multi-objective optimization framework named CMOMO.
• Implemented a dynamic constraint handling strategy to balance property optimization and constraint satisfaction.
• Utilized a latent vector fragmentation based evolutionary reproduction strategy for effective molecule generation.

Main Results:

• CMOMO outperformed five state-of-the-art methods on two benchmark tasks.
• The framework successfully generated optimized molecules with multiple desired properties and adherence to drug-like constraints.
• Validated on practical tasks, including protein-ligand optimization (4LDE) and inhibitor optimization (GSK3β), showing significant improvements.

Conclusions:

• CMOMO offers a superior approach to constrained molecular optimization compared to existing methods.
• The framework effectively balances multi-property optimization with critical drug-like constraints.
• Demonstrated significant success rate improvements, particularly for the GSK3β optimization task, highlighting its potential in drug discovery.