Efficient retrosynthetic planning with MCTS exploration enhanced A^* search

Area of Science:

• Computational Chemistry
• Organic Synthesis
• Artificial Intelligence in Chemistry

Background:

• Retrosynthetic planning is crucial for identifying synthetic routes in chemistry.
• Computer-aided retrosynthesis utilizes heuristic search algorithms like Monte Carlo Tree Search (MCTS) and A* search.
• Existing methods struggle with heuristic reliability, leading to search failures or suboptimal solutions due to inadequate or excessive exploration.

Purpose of the Study:

• To propose MCTS exploration enhanced A* (MEEA*) search for improved computer-aided retrosynthesis.
• To enhance the exploration capabilities of A* search by integrating MCTS's look-ahead functionality.
• To improve the generalization performance of guiding heuristics through path consistency regularization.

Main Methods:

• Development of the MEEA* algorithm, combining MCTS exploration with A* search.
• Implementation of a look-ahead search mechanism within the A* framework.
• Application of path consistency as a regularization technique for heuristic generalization.

Main Results:

• MEEA* demonstrated effectiveness across 10 diverse molecule datasets.
• Achieved a 100.0% success rate on the United States Patent and Trademark Office (USPTO) benchmark.
• Successfully identified bio-retrosynthetic pathways for 97.68% of natural product test compounds.

Conclusions:

• MEEA* significantly advances computer-aided retrosynthetic planning.
• The algorithm effectively balances exploration and exploitation for optimal pathway discovery.
• MEEA* offers a robust solution for complex retrosynthetic challenges in both general and natural product chemistry.