Back to today’s briefing

Story perspectives

RAG LLMs Excel in Disease-Gene Tasks, Need Better Knowledge Graphs

8/24/2025

30 2

Summary

Story summary

A study evaluated large language models (LLMs) on facial phenotypes, genes, and diseases using four query tasks.
Retrieval-Augmented Generation (RAG) LLMs outperformed standard LLMs in most tasks, effectively identifying fuzzy associations.
However, RAG LLMs faced challenges in coverage for phenotype-disease and phenotype-gene tasks.
The findings emphasize the necessity for better knowledge graphs to improve LLM performance.

1 of 1

Story summary

A study evaluated large language models (LLMs) on facial phenotypes, genes, and diseases using four query tasks.
Retrieval-Augmented Generation (RAG) LLMs outperformed standard LLMs in most tasks, effectively identifying fuzzy associations.
However, RAG LLMs faced challenges in coverage for phenotype-disease and phenotype-gene tasks.
The findings emphasize the necessity for better knowledge graphs to improve LLM performance.