Retrosynthetic planning with experience-guided Monte Carlo tree search.

Area of Science:

• Computational chemistry
• Organic synthesis
• Artificial intelligence in chemistry

Background:

• Retrosynthetic planning faces a combinatorial explosion of possibilities, challenging chemists in selecting optimal synthesis routes.
• Current methods for guiding retrosynthesis rely on limited chemical knowledge or costly estimation techniques.

Purpose of the Study:

• To develop a novel, experience-guided Monte Carlo tree search (EG-MCTS) algorithm for efficient and effective retrosynthetic planning.
• To address the limitations of existing score functions and estimation methods in complex molecule synthesis.

Main Methods:

• Implementation of an experience guidance network within the Monte Carlo tree search framework to learn from synthetic data.
• Utilizing synthetic experiences during the search process instead of traditional rollout methods.

Main Results:

• EG-MCTS demonstrated significant improvements in both efficiency and effectiveness compared to state-of-the-art approaches on USPTO datasets.
• Computer-generated synthesis routes closely matched reported routes in comparative literature experiments.
• Successful application of EG-MCTS in designing routes for real drug compounds, aiding chemists in retrosynthetic analysis.

Conclusions:

• EG-MCTS offers a powerful, experience-driven approach to overcome challenges in retrosynthetic planning.
• The developed method enhances the ability to discover optimal synthesis pathways for complex molecules.
• EG-MCTS shows strong potential for assisting chemists in practical drug discovery and development.