Presentation Information
[3J2-OS-15b-04]Development of an Interactive Electronic Lab Notebook for Chemical Research Based on AI Agents and a Knowledge Graph
〇Hinata Kadowaki1, Masahiko Nakano2,3, Junji Seino1,2 (1. School of Advanced Science and Engineering, Waseda University, 2. Waseda Research Institute for Science and Engineering, 3. SambaNova)
Keywords:
Large Language Model,Agentic AI,Knowledge Graph,Electronic Lab Notebook
Experimental knowledge in chemistry is often person-dependent, which limits its transferability. Existing electronic laboratory notebooks (ELNs) assume structured input formats, making it difficult to flexibly record changes in experimental operations in real time. This study proposes a novel conversational ELN framework based on large language model (LLM) agents and knowledge graphs (KGs) to support researchers throughout the experimental workflow.
The proposed system employs a multi-agent architecture to integrate information from literature, experimental data, and user inputs into a KG. During the preparation phase, experimental protocols are constructed through dialogue with agents, with operations and reagents represented as nodes in the graph. During the execution phase, operations and observations are interactively recorded and reflected in the KG. The constructed KG enables information retrieval and traceability after the experiment.
The system was validated through a demonstration use case involving the synthesis of a specific organic compound, constructing a KG from relevant literature and databases. During experimental execution, system response times ranged from a few seconds to approximately ten seconds, demonstrating the feasibility of context-aware interactive recording as well as real-time suggestions of alternative experimental plans.
The proposed system employs a multi-agent architecture to integrate information from literature, experimental data, and user inputs into a KG. During the preparation phase, experimental protocols are constructed through dialogue with agents, with operations and reagents represented as nodes in the graph. During the execution phase, operations and observations are interactively recorded and reflected in the KG. The constructed KG enables information retrieval and traceability after the experiment.
The system was validated through a demonstration use case involving the synthesis of a specific organic compound, constructing a KG from relevant literature and databases. During experimental execution, system response times ranged from a few seconds to approximately ten seconds, demonstrating the feasibility of context-aware interactive recording as well as real-time suggestions of alternative experimental plans.
Comment
To browse or post comments, you must log in.Log in