New Dataset and Method Extract Drug-Disease Applicability Conditions

Identifying the specific conditions under which a drug effectively treats a disease is crucial for informed clinical decision-making. However, most existing biomedical information extraction methods primarily focus on identifying drug-disease relationships, often overlooking these vital context-specific applicability conditions. To address this gap, researchers have introduced a new task: applicability condition extraction for therapeutic drug-disease relations from biomedical research literature. To facilitate this task, the team created the first dataset specifically designed for this purpose. This dataset features manually annotated triples of drugs, diseases, and their corresponding applicability conditions, derived from 1,119 drug-disease pairs found in biomedical paper abstracts. This comprehensive dataset provides a robust foundation for evaluating and developing new extraction methods. Using this new dataset, the researchers systematically evaluated a range of existing information extraction methods. Additionally, they proposed a novel method that enhances LoRA (Low-Rank Adaptation) to specifically consider the intricate relations between drugs and diseases, along with their conditions. Their proposed method consistently outperformed strong baselines across various evaluation settings, demonstrating its effectiveness in accurately extracting these critical applicability conditions.