CSearch: chemical space search via virtual synthesis and global optimization

Area of Science:

• Drug discovery and development
• Computational chemistry
• Medicinal chemistry

Background:

• Molecular design involves generating molecules and predicting their properties.
• Virtual screening and library screening are common but computationally intensive methods.

Purpose of the Study:

• To develop an effective method for molecular generation using virtual synthesis and global optimization.
• To optimize compounds for a specific objective function, such as binding affinity.

Main Methods:

• Utilized a pre-trained graph neural network (GNN) to approximate docking energies for four target receptors.
• Employed a global optimization algorithm with fragment-based virtual synthesis (CSearch method).

Main Results:

• Generated highly optimized compounds with 300-400 times less computational effort than virtual screening.
• Achieved similar synthesizability and diversity to known binders, with high potency and novelty.
• Produced drug-like binders with predicted binding poses similar to known inhibitors.

Conclusions:

• The CSearch method effectively explores chemical space for drug-like molecules.
• Generated compounds are optimized for objective functions, efficient, synthesizable, diverse, and novel.
• Demonstrated effectiveness in producing drug-like binders with favorable properties.