Improving Automated Deep Phenotyping with Large Language Models
Author Information
Author(s): Garcia Brandon T., Westerfield Lauren, Yelemali Priya, Gogate Nikhita, Andres Rivera-Munoz E., Du Haowei, Dawood Moez, Jolly Angad, Lupski James R., Posey Jennifer E.
Primary Institution: Baylor College of Medicine
Hypothesis
Can Retrieval-Augmented Generation improve the accuracy of phenotype extraction using large language models?
Conclusion
RAG-HPO significantly enhances the accuracy of phenotype extraction compared to traditional tools.
Supporting Evidence
- RAG-HPO achieved a mean precision of 0.84 and recall of 0.78.
- RAG-HPO identified twice as many correct HPO terms compared to other methods.
- False positive HPO term identification occurred for 15.8% of terms.
Takeaway
This study created a tool that helps computers better understand and identify genetic disorders by using advanced language models.
Methodology
The study developed a Python-based tool called RAG-HPO that uses a vector database and large language models to extract and assign HPO terms from clinical text.
Potential Biases
Subjectivity in clinical evaluations may affect phenotype interpretation.
Limitations
RAG-HPO's processing speed is slower than some existing tools, and its accuracy depends on the quality of the vector database.
Participant Demographics
The study analyzed 120 published case reports.
Statistical Information
P-Value
p<0.00001
Statistical Significance
p<0.00001
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website