Abstract
Clinical researchers need timely, evidence-supported overviews that clarify the underlying mechanisms and pathways of disease development. Manual curation of causal mechanisms is highly time-consuming and increasingly infeasible due to the exponential expansion of the biomedical literature. This paper introduces KnowDisease - an approach to automatically extracting disease theory mechanisms and causal relationships from biomedical literature using Large Language Models, exploiting Chain-of-Thought (CoT) prompting with evidence traceability and constructing Knowledge Graph-based disease theories for causal analysis. A proof-of-concept application is implemented based on this approach, enabling researchers to visualize complex disease pathways and causal relationships, navigate interconnected biological mechanisms and identify evidence for various disease theories based on constructed knowledge graphs. The initial results demonstrate how LLM-assisted tools can advance understanding of disease mechanisms and potentially accelerate biomedical research.