Molecular Graph Generation by Decomposition and Reassembling.

Area of Science:

• Computational chemistry and cheminformatics
• Drug discovery and materials science

Background:

• Designing molecules with specific properties is crucial but challenging due to vast chemical space.
• Current methods often struggle with interpretability and efficiency in exploring molecular candidates.

Purpose of the Study:

• To develop a novel, interpretable molecular generation method for drug discovery and materials design.
• To overcome the limitations of combinatorial explosion in identifying optimized molecular structures.

Main Methods:

• A two-step approach involving decomposition and reassembling of molecular structures.
• Frequent subgraph mining to identify molecular building blocks from a database.
• Reinforcement learning to guide the selection and combination of building blocks for new molecule generation.

Main Results:

• The proposed method successfully generates molecules with improved penalized log P and druglikeness scores.
• Experimental validation demonstrates the generation of valid intermediate molecules and drug candidates.
• The decomposition-and-reassembling process offers high interpretability without hidden space optimization.

Conclusions:

• The novel decomposition-and-reassembling method provides an effective and interpretable strategy for molecular design.
• This approach enhances the efficiency and quality of molecule generation in drug discovery and materials science.
• The method's ability to generate valid drug molecules with interpretable steps marks a significant advancement.