EpiGraph: A Knowledge Graph and Benchmark for Evidence-Intensive Reasoning in Epilepsy

Researchers have developed EpiGraph, a comprehensive knowledge graph containing 24,324 entities and 32,009 evidence-grounded triplets from 48,166 peer-reviewed papers to improve AI-driven epilepsy diagnosis and treatment. The accompanying EpiBench benchmark demonstrates that integrating structured clinical knowledge into large language models significantly enhances clinical reasoning, with improvements up to 41% in pharmacogenomic applications.

EpiGraph represents a significant advancement in applying structured knowledge graphs to specialized medical domains. The project addresses a critical gap in clinical AI by creating a systematically organized representation of epilepsy knowledge spanning biosignals, genetics, pharmacogenomics, and treatment strategies. This heterogeneous graph approach enables more precise clinical reasoning by grounding AI decisions in evidence-based medical literature.

The benchmark's evaluation of six LLMs under Graph-RAG settings demonstrates a practical methodology for measuring how knowledge augmentation impacts clinical decision-making. The dramatic 30-41% improvement in pharmacogenomic reasoning highlights the particular value of structured knowledge in precision medicine applications, where treatment outcomes depend on understanding complex drug-gene interactions.

For the medical AI industry, EpiGraph establishes both a technical framework and evaluation standard that could be replicated across other neurological conditions and medical specialties. The integration of multiple clinical layers—from raw biosignals to treatment outcomes—shows how comprehensive knowledge organization can bridge the gap between raw data and clinical application. This approach directly addresses one of healthcare AI's fundamental challenges: enabling models to reason across diverse, heterogeneous information sources.

The availability of code and benchmarks creates infrastructure for future research, potentially accelerating development of knowledge-augmented clinical AI systems. As healthcare organizations increasingly adopt AI for clinical support, having validated frameworks for knowledge integration becomes essential. The work validates that structured knowledge remains critical even in the era of large language models, suggesting hybrid approaches combining neural networks with explicit knowledge representation will define next-generation clinical AI.

• →EpiGraph integrates 48,166 peer-reviewed papers into a structured knowledge graph with 24,324 entities designed specifically for epilepsy clinical reasoning.
• →Graph-RAG augmentation improved LLM performance across all five clinical tasks, with pharmacogenomic reasoning showing the largest gains of 30-41%.
• →The benchmark framework evaluates six LLMs across five clinically motivated tasks including treatment recommendation and EEG report generation.
• →Structured knowledge integration addresses a critical limitation in current LLMs by grounding clinical decisions in organized evidence-based information.
• →The open-source release enables development of similar knowledge graphs for other medical specialties and neurological conditions.