eData: A tool and format for rapid extraction of exposure data from scientific articles

Abstract

Abstract Extracting and formatting data is among the most tedious elements of assessing environmental chemical occurrence and exposure. We present a toolkit (eData) to streamline this process, generate FAIR data, and maximise traceability and transparency: a FAIR Data Reporting Format (DRF) that provides data structure, vocabulary, and validation, and a Large Language Model (LLM)-based data extraction application that quickly formats literature data. Data and metadata for a study of interest can be rapidly extracted to the eData format by a range of commercially available LLMs, and visualised for checking in the application's graphical interface. This blends advantages of LLM extraction with human interpretation, speeding extraction without impairing reproducibility. Data can be exported from the application at any point in this process and saved locally. Furthermore, the application supports a streamlined Criteria for Reporting and Evaluating Exposure Datasets (CREED) data quality assessments that automatically compiles data relevant to the assessment for reporting. All resulting data can also be uploaded to the repository service Zenodo via a dedicated module. Our evaluation showed a substantial time saving per paper (from 3+ to around 1 hour) and extracted consistent data, passing 85% of format validation checks. Our toolkit is a valuable addition to the FAIRification of environmental chemical occurrence and exposure data.