MolSearch: Search-based Multi-objective Molecular Generation and Property Optimization.

Area of Science:

• Computational chemistry
• Drug discovery
• Molecular modeling

Background:

• Generating small molecules with specific properties is crucial for drug discovery.
• Simultaneously optimizing multiple molecular properties presents a significant challenge.
• Existing computational methods often struggle with efficiency and multi-objective optimization.

Purpose of the Study:

• To introduce MolSearch, a novel framework for multi-objective molecular generation.
• To demonstrate the effectiveness and computational efficiency of a search-based approach.
• To provide a viable alternative to deep learning methods in molecular generation.

Main Methods:

• Developed MolSearch, a search-based framework for molecular optimization.
• Employed a two-stage search strategy starting from existing molecules.
• Utilized systematically derived transformation rules from large compound libraries.

Main Results:

• MolSearch achieves performance comparable or superior to deep learning methods.
• The framework demonstrates high computational efficiency, enabling massive chemical space exploration.
• Effectiveness validated across multiple benchmark molecular generation settings.

Conclusions:

• Search-based methods, like MolSearch, are effective for multi-objective molecular generation.
• MolSearch offers a computationally efficient and powerful tool for drug discovery.
• The approach facilitates the exploration of vast chemical spaces under resource constraints.