Enhancing Monte Carlo Tree Search for Retrosynthesis.

Area of Science:

• Organic Chemistry
• Computational Chemistry
• Artificial Intelligence

Background:

• Computer-assisted synthesis programs commonly use neural networks and heuristic search algorithms.
• The Monte Carlo tree search (MCTS) algorithm is a key component in many such tools.
• Optimizing computational efficiency and route discovery is crucial for practical applications.

Purpose of the Study:

• To introduce two novel enhancements, eUCT and dUCT, to the MCTS algorithm for computer-assisted synthesis.
• To evaluate the impact of these enhancements on computational time and the number of synthetic routes found.
• To integrate these improvements into existing open-source tools and assess their performance on large molecule datasets.

Main Methods:

• Development and implementation of two MCTS enhancements: eUCT and dUCT.
• Integration of eUCT and dUCT into the AiZynthFinder program and the AI4Green electronic lab notebook.
• Utilizing a memory-efficient stock file to minimize computational carbon footprint.
• Testing the enhanced algorithms on sets of 1500 heavy molecules and 50,000 molecules from ChEMBL.

Main Results:

• Both eUCT and dUCT reduced computational clock-time by up to 50% for solving 1500 heavy molecules.
• The dUCT enhancement increased the number of synthetic routes discovered per molecule for both tested datasets.
• eUCT and dUCT-v2 solved 600-900 more molecules than unenhanced MCTS across the 50,000-molecule set.
• Under a 150s time constraint, dUCT-v1 found approximately 5 million more routes for the 50,000 targets compared to unenhanced MCTS.

Conclusions:

• The proposed eUCT and dUCT enhancements significantly improve the efficiency and route-finding capabilities of MCTS in computer-assisted synthesis.
• These advancements contribute to faster and more comprehensive exploration of synthetic possibilities for organic chemists.
• The integration into open-source platforms like AI4Green promotes wider adoption and sustainable computational chemistry practices.