This GitHub repository contains code created to analyze the National Library of Medicine CellLink corpus, as reported in the manuscript cited below. The analysis includes computations of corpus statistics, inter-annotator agreement (IAA), comparisons with other corpora, named entity recognition (NER) of cell populations, and entity linking (EL) of cell populations.

Repository Structure

This repository includes the following Python scripts, bash files, and other useful files. The cited figures and tables correspond to the manuscript cited below.

• paper_basic_stats.py – Computes basic statistics about the corpus (Table 1; Supplementary Tables 2, 5, and 6; Supplementary Figure 2)
• comparing_other_corpora_stats.py – Computes statistics comparing CellLink with other corpora that contain cell types (Table 2, Supplementary Table 7)
• NER-exp1-corpusComparisons/run_corpusComparisons.sh – Performs and analyzes NER experiments comparing models trained on CellLink or on other corpora that contain cell types (Table 3)
• NER-exp2-multipleModels/run_NER-exp2_BERT.sh and NER-exp2-multipleModels/run_NER-exp2_GPT.sh – Performs and analyzes NER experiments comparing different types of models trained on CellLink (Table 4)
• EL/run_EL.sh – Performs entity linking experiments (Table 5, Figure 4, Supplementary Tables 12 and 13)
• EL/EL_eval.py - Analyzes entity linking experiments (Table 5, Figure 4, Supplementary Tables 12 and 13)
• inter-annotator_agreement_evaluation.py – Computes inter-annotator agreement (Supplementary Tables 4 and 5)
• general_scripts/ – Shared utilities and helper scripts
• format_validation/ – Performs CellLink data format validation
• environments/ – Python environment specifications for BERT and GPT experiments
• Cell-Ontology_v2025-01-08.json – Cell Ontology version used for CellLink

Data Availability

The exact versions of the training and validation sets of the NLM CellLink corpus can be found here: . The test set is currently not released in full; further information can be found in the latest version of the same Zenodo data repository.

Citation

If you use this code or the CellLink corpus, please cite the following paper: Rotenberg N, Leaman R, Islamaj R, Kuivaniemi H, Tromp G, Fluharty B, Richardson S, Eastwood C, Diller M, Xu B, Pankajam A, Osumi-Sutherland D, Lu Z, & Scheuermann, R. H. Cell phenotypes in the biomedical literature: a systematic analysis and text mining corpus. bioRxiv. doi:10.64898/2026.02.11.705457.

References for other corpora containing cell types:

• AnatEM (Anatomical Entity Mention) corpus: Pyysalo S, Ananiadou S. Anatomical entity mention recognition at literature scale. Bioinformatics 30, 868-875 (2014).
• BioID (BioCreative VI BioID) corpus: Arighi C, et al. Bio-ID track overview. Proc. BioCreative Workshop 482, 376 (2017).
• CRAFT (Colorado Richly Annotated Full Text) corpus: Verspoor K, et al. A corpus of full-text journal articles is a robust evaluation tool for revealing differences in performance of biomedical natural language processing tools. BMC Bioinformatics 13, 207 (2012).
• JNLPBA (2004 Joint Workshop on Natural Language Processing in Biomedicine and its Applications) corpus: Collier N, Ohta T, Tsuruoka Y, Tateisi Y, Kim J-D. Introduction to the Bio-entity Recognition Task at JNLPBA. International Joint Workshop on Natural Language Processing in Biomedicine and its Applications (NLPBA/BioNLP), 73-78 (2004).