GEN-KnowRD: Reframing AI for Rare Disease Recognition.

Area of Science:

• Computational biology
• Medical informatics
• Genomics

Background:

• Rare diseases impact over 300 million globally, often leading to diagnostic delays.
• Current computational rare disease recognition (RDR) methods struggle with incomplete knowledge resources and scalability.
• Large language models (LLMs) for direct diagnosis face knowledge bottlenecks and practical concerns.

Purpose of the Study:

• To introduce GEN-KnowRD, a novel framework for building computable rare disease knowledge bases.
• To leverage LLMs for generating and assessing rare disease profiles.
• To develop lightweight inference pipelines for RDR using LLM-generated knowledge.

Main Methods:

• GEN-KnowRD framework utilizes LLMs to create schema-guided rare disease profiles.
• Quality assessment of generated profiles and construction of a computable knowledge base (PheMAP-RD).
• Integration of knowledge into lightweight inference pipelines for disease screening and early discrimination from EHRs.

Main Results:

• GEN-KnowRD significantly improved disease ranking on six public benchmarks (up to 345.8% top-1 success) compared to existing methods.
• Demonstrated robust discrimination performance gains in real-world cohorts for early idiopathic pulmonary fibrosis diagnosis.
• Outperformed state-of-the-art HPO-centered frameworks and end-to-end LLM reasoning approaches.

Conclusions:

• Repositioning LLMs to the knowledge layer enhances RDR performance.
• GEN-KnowRD provides a scalable, updatable, and reusable infrastructure for rare disease diagnosis and research.
• The framework supports improved screening and early discrimination, addressing critical needs in rare disease management.