Integrating Medicinal Chemist Expertise with Deep Learning for Automated Molecular Optimization.

Area of Science:

• Medicinal Chemistry
• Computational Chemistry
• Drug Discovery

Background:

• Medicinal chemist expertise is crucial for successful compound optimization.
• Existing molecular optimization strategies are often limited.
• A need exists for systematic, data-driven approaches to expand optimization strategies.

Purpose of the Study:

• To develop an automatic platform, AutoOptimizer, for molecular optimization.
• To create a novel framework, MolOpt, leveraging graph deep learning and expert knowledge.
• To build the first deep learning-generated molecular optimization strategy database.

Main Methods:

• Curated ~9000 molecular optimization strategies from literature.
• Developed the MolOpt framework using graph deep learning.
• Integrated expert knowledge and MolOpt into the AutoOptimizer platform.
• Conducted case studies on FGFR4 and HPK1 inhibitors.

Main Results:

• Identified potent inhibitors M8 and M9 against FGFR4 and HPK1.
• Achieved significant improvements: 77.6-fold for FGFR4 and 51.6-fold for HPK1.
• Demonstrated AutoOptimizer's practical application and effectiveness in lead optimization.

Conclusions:

• AutoOptimizer represents a novel, AI-driven approach to molecular optimization.
• The platform accelerates lead optimization and advances drug discovery.
• This work establishes a unique database grounded in medicinal chemistry expertise.