SynPlanner: An End-to-End Tool for Synthesis Planning

Area of Science:

• Computational Chemistry
• Chemoinformatics
• Organic Synthesis

Background:

• Retrosynthetic analysis is crucial for planning chemical synthesis.
• Developing customized retrosynthetic planners requires robust tools and methodologies.
• Existing tools may lack the flexibility needed for diverse synthetic challenges.

Purpose of the Study:

• To introduce SynPlanner, an end-to-end tool for designing customized retrosynthetic planners.
• To provide a flexible platform supporting various stages of retrosynthetic planning development.
• To enhance the efficiency and accessibility of creating bespoke retrosynthetic solutions.

Main Methods:

• A comprehensive reaction data curation pipeline including atom-to-atom mapping, standardization, and filtration.
• Reaction rule extraction and retrosynthetic model training.
• Retrosynthetic planning utilizing Monte Carlo Tree Search (MCTS) guided by graph neural networks.

Main Results:

• SynPlanner offers a customizable pipeline for building tailored retrosynthetic planning tools.
• The tool integrates reaction data processing, rule extraction, model training, and planning.
• Graph neural networks guide MCTS for accurate retrosynthetic rule predictions and precursor synthesizability assessment.

Conclusions:

• SynPlanner provides a flexible and comprehensive solution for developing customized retrosynthetic planners.
• The tool's modular design allows adaptation to specific research and development needs.
• Accessible via GUI and CLI, SynPlanner is supported by tutorials and available on GitHub.