KRAGEN: a knowledge graph-enhanced RAG framework for biomedical problem solving using large language models

• DOI: 10.1093/bioinformatics/btae353
• Publication Date: 2024-06-03T20:52:45Z
• Authors: Nicholas Matsumoto; Jay Moran; Hyunjun Choi; Miguel E Hernandez; Mythreye Venkatesan; Paul Wang; Jason H Moore
• Keywords: Biomedicine

Abstract Motivation Answering and solving complex problems using a large language model (LLM) given certain domain such as biomedicine is challenging task that requires both factual consistency logic, LLMs often suffer from some major limitations, hallucinating false or irrelevant information, being influenced by noisy data. These issues can compromise the trustworthiness, accuracy, compliance of LLM-generated text insights. Results Knowledge Retrieval Augmented Generation ENgine (KRAGEN) new tool combines knowledge graphs, (RAG), advanced prompting techniques to solve with natural language. KRAGEN converts graphs into vector database uses RAG retrieve relevant facts it. techniques: namely graph-of-thoughts (GoT), dynamically break down problem smaller subproblems, proceeds each subproblem through framework, which limits hallucinations, finally, consolidates subproblems provides solution. KRAGEN’s graph visualization allows user interact evaluate quality solution’s GoT structure logic. Availability implementation deployed running its custom Docker containers. available open-source GitHub at: https://github.com/EpistasisLab/KRAGEN.