CanRisk-RAG Delivers Transparent Cancer Model Recommendations

Researchers at Sichuan University publish CanRisk-RAG, a retrieval-augmented, knowledge-guided platform for recommending cancer risk prediction models. Built on a curated knowledge base of more than 800 peer-reviewed models, it uses LLM-based semantic tagging, embeddings, multifactor ranking, and LLM summaries to produce structured recommendations. On an independent validation set and expert review, CanRisk-RAG outperformed PubMed, ChatGPT-4o, ScholarAI, and Gemini 1.5 Flash, scoring 8.30 relevance and 7.62 reliability, though broader clinical validation is still needed.