RetroPath RL

RetroPath RL applies Monte Carlo Tree Search reinforcement learning to explore bioretrosynthesis pathways and propose enzymatic synthesis plans for metabolic engineering.

Key Features:

• Monte Carlo Tree Search Reinforcement Learning: Uses Monte Carlo Tree Search (MCTS) reinforcement learning to explore and rank candidate bioretrosynthesis routes.
• Chemical similarity guidance: Guides the MCTS search using chemical similarity metrics to prioritize chemically plausible transformations.
• Compatibility with Mono-component Reaction Rules: Operates with mono-component reaction rules from RetroRules.
• Media supplement suggestions: Suggests potential media supplements to complement enzymatic synthesis plans.
• Validation datasets: Validated on a "golden" dataset of 20 manually curated experimental pathways and a dataset of 152 successful metabolic engineering projects.

Scientific Applications:

• Metabolic engineering pathway design: Designs biosynthetic pathways for production of chemicals via engineered organisms.
• Enzymatic synthesis planning: Generates and evaluates enzymatic synthesis plans for candidate target compounds.

Methodology:

Applies Monte Carlo Tree Search reinforcement learning guided by chemical similarity and uses mono-component reaction rules from RetroRules.