(59as) Prediction of Chemical Toxicity and Exposure Symptoms Based on Knowledge Graph Embedding and Language Models

If the initial response to a chemical exposure accident is inadequate, there is a possibility of significant human and property damage. Research on a knowledge service for hazardous chemicals that considers detection and symptoms in a dynamically changing environment is currently insufficient. Therefore, there is a need for real-time hazardous chemical detection and diagnostic systems and knowledge services that can quickly detect exposed chemicals and respond to hazardous situations, protecting workers or exposed individuals from unexpected chemical accidents on site. In this study, an AI-based system for chemical identification was developed based on actively collected exposure symptoms for real-time response to hazardous material accidents. To analyze chemical exposure accidents or substance contact symptoms, chemical-related information (symptoms, toxicity, structural information, etc.) is collected and extracted from WISER, PubChem, MSDS, and other sources to build a knowledge base. Compared to the method of using NIH WISER, which manually inputs symptoms to estimate chemicals, the proposed system actively collects symptoms and predicts candidate chemicals in probabilistic form through inference using the knowledge base.