Evidence-based Knowledge Synthesis and Hypothesis Validation: Navigating Biomedical Knowledge Bases via Explainable...

Area of Science:

• Biomedical Informatics
• Artificial Intelligence in Medicine
• Knowledge Discovery

Background:

• The vast scale of biomedical data presents challenges for researchers.
• Large Language Models (LLMs) offer potential but suffer from hallucinations.
• Retrieval-Augmented Generation (RAG) is crucial for enhancing LLM accuracy.

Purpose of the Study:

• Introduce RUGGED (Retrieval Under Graph-Guided Explainable disease Distinction), a workflow for knowledge integration and bias mitigation.
• Facilitate exploration and validation of new research directions in biomedicine.
• Enhance the reliability of LLM-driven biomedical research.

Main Methods:

• Synthesize biomedical information from literature and knowledge bases.
• Employ text-mining association analysis and explainable graph prediction models.
• Utilize RAG-enhanced LLMs for hypothesis exploration and mechanism investigation.

Main Results:

• Uncover potential drug-disease relationships and molecular interactions.
• Demonstrate RUGGED's utility in evaluating therapeutics for Arrhythmogenic Cardiomyopathy (ACM) and Dilated Cardiomyopathy (DCM).
• Reduce LLM hallucinations and highlight actionable insights for novel therapeutics.

Conclusions:

• RUGGED streamlines the investigation of novel therapeutics by integrating diverse biomedical data.
• The workflow enhances the accuracy and reliability of LLM applications in biomedical research.
• RUGGED supports informed decision-making for drug discovery and repositioning.