AutoPCR: automated phenotype concept recognition by prompting.
Yicheng Tao1, Yuanhao Huang2, Yiqun Wang2
1Department of Electrical Engineering and Computer Science, University of Michigan, Ann Arbor, MI 48109, United States.
Bioinformatics (Oxford, England)
|July 7, 2026
Summary
AutoPCR, a new method for phenotype concept recognition, automatically generalizes to new ontologies and data without specific training. This approach enhances biomedical text mining by improving performance and adaptability across diverse datasets.
Area of Science:
- Biomedical text mining
- Natural Language Processing
Background:
- Phenotype concept recognition (CR) is crucial for biomedical text mining.
- Existing CR methods face challenges in generalization due to ontology-specific training or lack of domain knowledge in general LLMs.
Purpose of the Study:
- To develop a novel prompt-based phenotype CR method, AutoPCR, that generalizes across diverse text styles and evolving terminology.
- To improve the adaptability and performance of phenotype concept recognition without requiring ontology-specific training.
Main Methods:
- Proposed AutoPCR, a prompt-based method for phenotype concept recognition.
- Introduced an optional self-supervised training strategy to further enhance performance.
- Evaluated AutoPCR's generalizability and inductive capabilities through ablation and transfer studies.
Main Results:
- AutoPCR achieved superior and more robust performance across various datasets compared to existing methods.
- Demonstrated significant inductive capability and generalizability to new ontologies.
- The method successfully addresses limitations of ontology-specific training and general-purpose LLMs.
Conclusions:
- AutoPCR offers a generalizable and effective solution for phenotype concept recognition in biomedical text mining.
- The proposed method advances the field by enabling adaptation to new ontologies and unseen data.
- AutoPCR provides a robust framework for phenotype CR, enhancing biomedical data analysis.
