GEN-KnowRD: Reframing AI for Rare Disease Recognition
Yan, C.; Su, W.-C.; Xin, Y.; Grabowska, M. E.; Kerchberger, V. E.; Borza, V. A.; Wang, J.; Wang, L.; Li, R.; Lynn, J.; Dickson, A. L.; Shyr, C.; Feng, Q.; Stein, C. M.; Wang, K.; Embi, P.; Malin, B. A.; Liu, H.; Wei, W.-Q.
Show abstract
Rare diseases affect over 300 million people worldwide, yet patients often endure years-long diagnostic delays that limit timely intervention and trial opportunities. Computational rare disease recognition (RDR) remains constrained by knowledge resources that are often incomplete, heterogeneous, and dependent on extensive multi-disciplinary expert curation that cannot scale. Large language models (LLMs) applied directly for end-to-end diagnosis or disease discrimination face similar knowledge bottlenecks while also raising concerns around cost, reproducibility, and data governance. Here, we introduce GEN-KnowRD, a knowledge-layer-first framework that leverages LLMs to generate schema-guided rare disease profiles, systematically assesses their quality, and constructs a computable knowledge base (PheMAP-RD) for local deployment. GEN-KnowRD integrates this knowledge into lightweight inference pipelines for both general-purpose disease screening and specialized early discrimination from longitudinal electronic health records. Across six public benchmarks for general-purpose screen (9,290 patients spanning 798 rare diseases), GEN-KnowRD significantly improves disease ranking compared to a state-of-the-art, HPO-centered diagnostic framework (up to 345.8% improvement in top-1 success), advanced end-to-end LLM reasoning (up to 129.1% improvement), and a variant of GEN-KnowRD instantiated with expert-curated knowledge rather than LLM-generated profiles. In two real-world cohorts for early diagnosis of idiopathic pulmonary fibrosis (511 patients) as a use case, GEN-KnowRD also demonstrates robust discrimination performance gains, supporting effective RDR during the pre-diagnostic window. These findings demonstrate that repositioning LLMs from diagnostic reasoning to the knowledge layer--decoupling knowledge construction from patient-level inference--yields stronger RDR, while providing scalable, continuously updatable, and reusable infrastructure for diagnosis, screening, and clinical research across the rare disease landscape.
Matching journals
The top 3 journals account for 50% of the predicted probability mass.
Similar papers in this journal
- Few shot learning for phenotype-driven diagnosis of patients with rare genetic diseases 97%
- Clinical Knowledge Extraction via Sparse Embedding Regression (KESER) with Multi-Center Large Scale Electronic Health Record Data 95%
- Large language models improve transferability of electronic health record-based predictions across countries and coding systems 95%
Similar papers in this journal
Similar papers in this journal
- Pretrained Patient Trajectories for Adverse Drug Event Prediction Using Common Data Model-based Electronic Health Records 94%
- The Interpretable Multimodal Machine Learning (IMML) framework reveals pathological signatures of distal sensorimotor polyneuropathy 92%
- Leveraging sequences missing from the human genome to diagnose cancer 92%
Similar papers in this journal
- Transformer-based deep learning model for the diagnosis of suspected lung cancer in primary care based on electronic health record data 94%
- Consistent Performance of GPT-4o in Rare Disease Diagnosis Across Nine Languages and 4967 Cases 93%
- Integrative deep learning analysis improves colon adenocarcinoma patient stratification at risk for mortality 91%
"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.