The abstract is read at the publisher; the summary is JClub's.
A modular biomedical named entity normalization workflow for HPO, CL, Uberon, and GO shows retrieval and reranking performance that varies by ontology and benchmark, with larger candidate sets improving recall but not necessarily normalization accuracy.
- Why it matters: Rare disease mechanism studies depend on integrating phenotypes, cell types, anatomy, and molecular functions, but relevant relationships are dispersed in literature and require reliable normalization across specialized ontologies.
- What they did: The workflow combines exact matching, lexical and semantic candidate retrieval (BM25, fuzzy matching, character n-grams, SapBERT), and reranking, evaluated across multiple benchmarks with separate scoring for retrieval, reranking, and overall normalization accuracy.
- The result: Candidate-set size increases retrieval recall but can reduce reranking performance, and retrieval difficulty is linked to mention characteristics and ontology categories, informing where the approach is limited and how to assess components independently.
The findingWhy it mattersWhat they didThe result
AI summary of the abstract · may contain errors · not medical advice · read the paper.