A modular workflow for biomedical entity normalization across heterogeneous biological ontologies

Understanding the biological mechanisms underlying rare diseases requires integrating information about phenotypes, cell types, anatomical structures, and molecular functions and processes. Although some of this knowledge is available in structured resources, additional relationships remain dispersed across biomedical literature. Extracting and integrating this information requires normalizing heterogeneous biological entities against specialized ontologies. This motivated the development of a common biomedical named entity normalization (NEN) workflow and the evaluation of its performance across different entity types and ontologies. We developed a modular NEN workflow combining exact matching, lexical and semantic candidate retrieval, and reranking, without requiring ontology-specific training. Candidate retrieval integrates BM25, fuzzy matching, character n-gram similarity, and SapBERT. We evaluated the workflow across the Human Phenotype Ontology (HPO), Cell Ontology (CL), Uberon, and Gene Ontology (GO) using multiple benchmarks, separately assessing candidate retrieval, reranking accuracy, and overall normalization accuracy. We also investigated factors associated with retrieval difficulty and compared the workflow with CellExLink, a specialized cell-type normalization system. Retrieval performance varied across ontologies and benchmarks, with lexical and semantic methods providing complementary candidate coverage in some settings. Increasing candidate-set size improved retrieval recall but did not necessarily improve overall normalization accuracy, as larger candidate sets could reduce reranking performance. Retrieval difficulty was associated with mention characteristics and ontology categories. Our evaluation identifies the performance characteristics and limitations of a common normalization workflow across heterogeneous biological ontologies, highlighting the importance of assessing candidate retrieval and selection separately. The workflow provides a methodological foundation for extracting and integrating biological information from rare disease literature, with potential applicability to other biomedical information-extraction tasks requiring normalization against these ontologies.

Authors

Institutions

Publication Details

Journal
bioRxiv (Cold Spring Harbor Laboratory)
Published
2026-11-19
DOI
https://doi.org/10.64898/2026.11.17.767229
Primary Topic
Biomedical Text Mining and Ontologies
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

A modular workflow for biomedical entity normalization across heterogeneous biological ontologies

Mónica Chagoyen, Ana Amil
bioRxiv (Cold Spring Harbor Laboratory)
Biomedical Text Mining and Ontologies
preprint

A modular workflow for biomedical entity normalization across heterogeneous biological ontologies

Mónica Chagoyen, Ana Amil
preprint en

Abstract

Understanding the biological mechanisms underlying rare diseases requires integrating information about phenotypes, cell types, anatomical structures, and molecular functions and processes. Although some of this knowledge is available in structured resources, additional relationships remain dispersed across biomedical literature. Extracting and integrating this information requires normalizing heterogeneous biological entities against specialized ontologies. This motivated the development of a common biomedical named entity normalization (NEN) workflow and the evaluation of its performance across different entity types and ontologies. We developed a modular NEN workflow combining exact matching, lexical and semantic candidate retrieval, and reranking, without requiring ontology-specific training. Candidate retrieval integrates BM25, fuzzy matching, character n-gram similarity, and SapBERT. We evaluated the workflow across the Human Phenotype Ontology (HPO), Cell Ontology (CL), Uberon, and Gene Ontology (GO) using multiple benchmarks, separately assessing candidate retrieval, reranking accuracy, and overall normalization accuracy. We also investigated factors associated with retrieval difficulty and compared the workflow with CellExLink, a specialized cell-type normalization system. Retrieval performance varied across ontologies and benchmarks, with lexical and semantic methods providing complementary candidate coverage in some settings. Increasing candidate-set size improved retrieval recall but did not necessarily improve overall normalization accuracy, as larger candidate sets could reduce reranking performance. Retrieval difficulty was associated with mention characteristics and ontology categories. Our evaluation identifies the performance characteristics and limitations of a common normalization workflow across heterogeneous biological ontologies, highlighting the importance of assessing candidate retrieval and selection separately. The workflow provides a methodological foundation for extracting and integrating biological information from rare disease literature, with potential applicability to other biomedical information-extraction tasks requiring normalization against these ontologies.

bioRxiv (Cold Spring Harbor Laboratory)
Instituto Cajal (ES)
Biomedical Text Mining and Ontologies
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon