GDReCo: Fine-grained gene-disease relationship extraction corpus

BACKGROUND AND OBJECTIVE: Understanding gene-disease relationships is crucial for medical research, drug discovery, clinical diagnosis, and other fields. However, there is currently no high-quality, fine-grained corpus available for training Natural Language Processing (NLP) models, which have proven to be effective in knowledge extraction. METHODS: This study introduces a novel ontology framework for gene-disease associations, addressing the absence of a formal descriptive system and training corpus for NLP models. RESULTS: We developed the Gene Disease Relationship Extraction Corpus (GDReCo), a refined dataset of over 24,000+ cases, including 2300+ manually annotated and 22,000+ model-predicted instances. BERT-based models trained on this data achieved high F1-scores for "event" and "rel" relationships, validating its effectiveness for Gene-Disease Relationship Extraction (GDRE) tasks. CONCLUSIONS: GDReCo serves as a valuable resource for biomedical research, though ChatGPT's limitations in fine-grained relation extraction are noted.