Retrieval-Augmented Generative Agent for Reaction Condition Recommendation in Chemical Synthesis

Recent artificial intelligence (AI) research plots a promising future of automatic chemical reactions within the chemistry society. This study presents a transformative AI agent that automates the reaction condition recommendation (RCR) task in chemistry using retrieval-augmented generation (RAG) technology. By emulating expert chemists search and analysis strategies, the agent employs large language models (LLMs) to interrogate molecular databases and distill critical data from online literature. Further, the AI agent is equipped with our novel reaction fingerprint developed for the RCR task. Thanks to the RAG technology, our agent uses updated online databases as knowledge sources, significantly outperforming conventional AIs confined to the fixed knowledge within its training data. The resulting system can significantly reduce chemists workload, allowing them to focus on more fundamental and creative scientific problems. This significant advancement brings closer computational techniques and chemical research, marking a considerable leap toward harnessing AI's full capabilities in chemical discovery.