This paper presents a domain-specific adaptation of GraphRAG framework for nutrigenetics, focusing on the extraction of genetic variant information relevant to personalized nutrition. By integrating Knowledge Graphs with Retrieval-Augmented Generation (RAG), we enhance biomedical knowledge discovery. A model selection study identifies optimal combinations of Large Language Models (LLMs) and embeddings, with Gemma2:9B paired with BERT achieving the highest quality score in graph construction. Evaluation against the naive RAG baseline shows significant improvements in response comprehensiveness, directness, and empowerment across diverse user profiles, including researchers, healthcare professionals, and consumers. These advancements highlight the potential of GraphRAG to accelerate hypothesis generation, support clinical decision-making, and empower individuals to make informed dietary choices based on genetic insights. These preliminary findings underscore the importance of structured knowledge representation in addressing biomedical challenges, with promising implications for advancing personalized nutrition and multi-domain biomedical applications.
Knowledge Graph-Enhanced Retrieval-Augmented Generation for Nutrigenetics / De Filippis, G.M., Benfenati, D., De Carlo, G., Rinaldi, A.M.. - 2696:(2026), pp. 91-104. (1st International Workshop on Artificial Intelligence for Biomedical Data, AIBio 2025, held in conjunction with the 28th European Conference on Artificial Intelligence, ECAI 2025 1st International Workshop on Artificial Intelligence for Biomedical Data, AIBio 2025, held in conjunction with the 28th European Conference on Artificial Intelligence, ECAI 2025 ita 2025) [10.1007/978-3-032-17216-7_8].
Knowledge Graph-Enhanced Retrieval-Augmented Generation for Nutrigenetics
Giovanni Maria De Filippis;Domenico Benfenati
;Antonio Maria Rinaldi
2026
Abstract
This paper presents a domain-specific adaptation of GraphRAG framework for nutrigenetics, focusing on the extraction of genetic variant information relevant to personalized nutrition. By integrating Knowledge Graphs with Retrieval-Augmented Generation (RAG), we enhance biomedical knowledge discovery. A model selection study identifies optimal combinations of Large Language Models (LLMs) and embeddings, with Gemma2:9B paired with BERT achieving the highest quality score in graph construction. Evaluation against the naive RAG baseline shows significant improvements in response comprehensiveness, directness, and empowerment across diverse user profiles, including researchers, healthcare professionals, and consumers. These advancements highlight the potential of GraphRAG to accelerate hypothesis generation, support clinical decision-making, and empower individuals to make informed dietary choices based on genetic insights. These preliminary findings underscore the importance of structured knowledge representation in addressing biomedical challenges, with promising implications for advancing personalized nutrition and multi-domain biomedical applications.| File | Dimensione | Formato | |
|---|---|---|---|
|
978-3-032-17216-7_8.pdf
accesso aperto
Licenza:
Copyright dell'editore
Dimensione
1.07 MB
Formato
Adobe PDF
|
1.07 MB | Adobe PDF | Visualizza/Apri |
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.


