PhenoAlign: A Hybrid Data-Knowledge-Driven Approach for Precisely Aligning Phenotype Information in Medical Texts

Area of Science:

• Medical Informatics
• Natural Language Processing
• Biomedical Data Science

Background:

• Accurate alignment of phenotypic information in medical texts is crucial for intelligent medical applications like patient case retrieval.
• A dedicated algorithm for this precise alignment task was previously lacking.

Purpose of the Study:

• To explore and identify optimal strategies for aligning the semantic structured unit of phenotypes (PhenoSSU).
• To develop a novel tool, PhenoAlign, for precise phenotype alignment in medical texts.

Main Methods:

• Investigated various PhenoSSU alignment strategies, focusing on a data-knowledge-driven approach.
• Utilized a BERT-based pre-trained language model for phrase-type PhenoSSU alignment.
• Employed a knowledge-based method for logic-type PhenoSSU alignment.
• Integrated PhenoSSU extraction and alignment algorithms into the PhenoAlign tool.

Main Results:

• The data-knowledge-driven approach proved most effective for PhenoSSU alignment.
• PhenoAlign achieved an end-to-end F1 score of 0.820 on an expert-annotated test set.
• PhenoSSU extraction and alignment F1 scores were 0.885 and 0.927, respectively.
• Identified significant challenges for large language models like ChatGPT in this domain.

Conclusions:

• PhenoAlign is a simple, effective tool for medical text phenotype alignment.
• The developed tool facilitates precise phenotypic information alignment, benefiting intelligent medical applications.
• This advancement supports patient care and medical research by improving data accessibility and interpretability.