Synergizing a knowledge graph and large language model for relay catalysis pathway recommendation.

Area of Science:

• Catalysis
• Chemical Engineering
• Computational Chemistry

Background:

• Relay catalysis enables efficient multistep transformations but designing pathways is labor-intensive.
• Current methods rely heavily on expert literature analysis, limiting discovery speed and scope.

Purpose of the Study:

• To develop an automated approach for recommending multistep relay catalysis pathways.
• To accelerate the design and discovery of novel catalytic reaction sequences.

Main Methods:

• Integration of a knowledge graph (KG) and large language models (LLMs) for pathway recommendation.
• LLM-assisted data acquisition and organization to build a comprehensive catalysis knowledge graph (Cat-KG).
• Scoring rules and LLM-driven text generation for pathway validation and presentation.

Main Results:

• The method successfully recommended relay catalysis pathways for ethylene, ethanol, and 2,5-furandicarboxylate within minutes.
• Identified pathways were consistent with reported ones, demonstrating effectiveness and potential for novel discoveries.
• The system generated readable chemical equations and descriptions, integrating reliable catalysis knowledge.

Conclusions:

• The combined KG and LLM strategy automates the discovery of relay catalysis pathways.
• This approach significantly reduces the time and cost associated with designing complex catalytic reactions.
• The strategy shows potential for extrapolating known and identifying novel relay catalysis pathways.