Tree-Invent: A Novel Multipurpose Molecular Generative Model Constrained with a Topological Tree.

Area of Science:

• Computational chemistry
• Medicinal chemistry
• Drug discovery

Background:

• De novo molecular design is crucial for identifying novel drug candidates.
• Existing generative models may lack precise control over molecular topology.

Purpose of the Study:

• To propose a novel generative model, Tree-Invent, for de novo molecular design.
• To integrate topological constraints for enhanced control in molecular graph generation.

Main Methods:

• Representing molecular graphs as topological trees with nodes for ring systems and atoms, and edges for bonds.
• Utilizing three submodels for node addition, ring generation, and node connection.
• Integrating reinforcement learning for efficient chemical space exploration.

Main Results:

• Tree-Invent successfully generates molecular graphs by incorporating topological constraints.
• The model allows for precise control over structure generation via specified topological trees.
• Demonstrated flexibility in various molecule design tasks, including scaffold decoration and hopping.

Conclusions:

• Tree-Invent offers a novel approach to de novo molecular design with integrated topological control.
• The model enhances the efficiency of exploring targeted chemical spaces in drug discovery.
• Tree-Invent provides a flexible and powerful tool for diverse medicinal chemistry applications.