Data and results for active learning-based text mining for chondrogenesis regulators with PubMedBERT and open weight LLMs

This dataset accompanies the GitHub repository ChondroTextomics/ALRegulatorDiscovery. It contains the data and outputs of a text-mining pipeline that identifies candidate regulators of chondrogenesis from PubMed abstracts, using a PubMedBERT classifier trained through active learning and benchmarked against a logistic regression baseline and open-weight large language models. It includes the annotated training data, per-curator and aggregated annotation files, the PMIDs needed to retrieve the source abstracts (abstract text is not redistributed), model predictions and evaluation results, and the scripts used. Model weights are available on Hugging Face Hub at amav/pubmedbert-chondrogenesis-classifier.