Pharmacokinetic Table Cell-Type Classification and Named Entity Recognition (PKTab-CTC-NER) Corpora

This dataset contains expert-annotated corpora for table cell type classification (CTC), pharmacokinetic named entity recognition (PK NER), and contextual entity recognition in pharmacokinetic (PK) literature tables. The corpora were developed as part of a unified annotation workflow for automated PK information extraction from biomedical tables. The corpora are provided in JSONL format with predefined development, validation, and held-out test splits where appropriate. The CTC corpus contains one record per non-numeric table cell extracted from full-text PK publications. Each record includes the cell text together with structural table features, including row and column headers, local row and column context, table region, and positional information. Cells are annotated as either PK, indicating that the cell contains an in vivo pharmacokinetic parameter, or Other, encompassing contextual information, covariates, variability terms, compounds, doses, and other non-PK table content. The PK NER corpus contains expert annotations of pharmacokinetic parameter mentions within PK-labelled table cells. Each record includes the cell text together with character-level span annotations for PK parameter mentions, source table identifiers, and contextual information from the original table. PK parameter spans were annotated using the complete table, caption, and footer as context. The annotation task focused on identifying in vivo PK parameters representing reported parameter estimates or values. The release also includes a pool of annotated non-PK table cells together with a script for reproducing the negative-example augmentation experiments (10%, 30%, 50%, and 100% negative examples) used during PK NER training and reported in the accompanying publication. The contextual entity corpus contains expert annotations of experimental information associated with PK parameters, including entities such as dose, units, number of subjects, biological sample, sampling time, administration route, measurement type, age, and weight. These annotations support contextual information extraction from PK tables and complement the PK parameter annotations. Please refer to the README.md file for further details. Together, these corpora support the development and evaluation of machine learning and rule-based methods for pharmacokinetic table understanding, cell type classification, named entity recognition, and contextual information extraction, forming the recognition component of the end-to-end PK information extraction pipeline described in the accompanying publication.