license: cc-by-sa-4.0
language:
- en
Dataset Card: OTAR3088 Cellfinder (Cleaned & Filtered)
This dataset is a curated derivative of the CellFinder corpus, specifically adapted for the OTAR3088 project. It focuses on high-precision entity recognition for cellular and tissue-based nomenclature.
Dataset Description
While the original CellFinder dataset provides 6(anatomical parts, cell components, cell lines, cell types, genes/protein and species) entity types, this version has been adapted and restructured to focus on three core biological entities: CellLine, CellType, and Anatomy(equivalent to Tissue entity in our project). This version also resolves some annotation overlaps and inconsistencies found in the source material to improve training performance in Transformer-based models.
The original Cellfinder annotated texts was sourced from https://www.informatik.hu-berlin.de/de/forschung/gebiete/wbi/resources/cellfinder/
Filters applied:
- Removal of "Anatomy" entities which are not fit for our use of "Anatomy" terms as "Tissue" equivalents
- Resolution of cases where a term is given multiple labels in data sets, resolved by choosing the mode label / failing that a list of priorities is used for the entity label that is given
- Drop entries of annotations which are outside of the entity types we care for, which are: "CellType", "CellLine" and "Anatomy"~Tissue
Dataset info
- Curated by: Christine Withers
- Funded by [optional]: Open Targets
- Shared by [optional]: [More Information Needed]
- Language(s) (NLP): English
Licensing & Attribution
License: CC BY-SA 4.0
License Note : This dataset is a derivative work based on the original CellFinder corpus (licensed under CC BY-SA 3.0). As permitted by the "Later Version" clause of the 3.0 license, this version is distributed under CC BY-SA 4.0.
Original Source: CellFinder - HU Berlin
Original Paper: Neves M, et al. CellFinder: a corpus for multiclass annotations of cellular components and molecules in biomedical texts-> PubMed
Dataset Structure
- BRAT