Graph Polish: A Novel Graph Generation Paradigm for Molecular Optimization.

Area of Science:

• Computational Chemistry
• Medicinal Chemistry
• Drug Discovery

Background:

• Molecular optimization is crucial for designing drug candidates with desired properties.
• Traditional methods often lack efficiency or fail to leverage supervised learning from optimized molecule pairs.

Purpose of the Study:

• Introduce Graph Polish, a novel paradigm for molecular optimization.
• Develop an efficient learning framework, Teacher and Student polish, to guide optimization steps.

Main Methods:

• Graph Polish predicts optimization centers within molecules and refines surrounding regions.
• The Teacher and Student polish framework uses a teacher component for annotation and a student component for learning and application.
• The paradigm offers intuitive interpretations of molecular optimization outcomes.

Main Results:

• The proposed Graph Polish approach demonstrated significant advantages over six state-of-the-art baseline methods.
• Experiments on benchmark datasets validated the effectiveness and explainability of the novel paradigm.
• The method also showed considerable time savings compared to existing techniques.

Conclusions:

• Graph Polish provides an effective, explainable, and efficient solution for molecular optimization in drug discovery.
• The Teacher and Student polish framework successfully captures dependencies in optimization steps.
• This paradigm represents a significant advancement in computational drug design.