Cell phenotypes in the biomedical literature: a systematic analysis and text mining corpus

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Single-cell technologies rapidly expand cell phenotype identification.
• Existing knowledge is fragmented across literature and lacks structured resources.
• Accurate cell population annotation is crucial for biological research.

Purpose of the Study:

• To develop the CellLink corpus, a manually annotated dataset of human and mouse cell populations.
• To link cell mentions to Cell Ontology (CL) terms, distinguishing specific, heterogeneous, and vague populations.
• To analyze patterns in cell naming and improve computational tools for cell type recognition and linking.

Main Methods:

• Manual annotation of over 22,000 cell population mentions from recent journal articles.
• Linking cell mentions to Cell Ontology (CL) terms as exact or related matches.
• Systematic analysis of author-used attributes (anatomical context, molecular signatures, etc.) in cell naming.
• Fine-tuning transformer-based models and using embedding-based approaches for named entity recognition and entity linking.

Main Results:

• The CellLink corpus covers nearly half of the current Cell Ontology terms.
• Lineage-specific patterns in cell naming attributes were identified.
• Fine-tuning models on CellLink significantly improved named entity recognition performance.
• Embedding-based methods demonstrated effectiveness in zero-shot entity linking and distinguishing match types.

Conclusions:

• CellLink is a valuable resource for standardizing cell population terminology and improving computational analysis.
• The corpus facilitates a deeper understanding of how cell phenotypes are described in scientific literature.
• CellLink aids in expanding and refining biological ontologies like the Cell Ontology.