PHAEDRA

PHAEDRA provides a semantically annotated corpus of MEDLINE abstracts to support pharmacovigilance text mining and machine-learning model development for extracting drug effects, drug-drug interactions, and adverse events.

Key Features:

• Rich Semantic Annotations: Multi-level semantic annotations capture drug effects, drug-drug interactions, adverse events, and relationships between drugs and diseases using comprehensive annotation guidelines.
• High Inter-Annotator Agreement: Annotations were validated with F-score 92.6% for named entities and 78.4% for complex events under relaxed matching criteria.
• Domain-Specific Adaptation: A corpus of 597 MEDLINE abstracts supports training and adaptation of text-mining and machine-learning models for pharmacovigilance-specific named entity recognition and event extraction.
• Evidence Identification: Encodes complex semantic information enabling identification of documents containing evidence about drug usage, effects, and patient-specific factors relevant to pharmacovigilance curation.

Scientific Applications:

• Automate Information Extraction: Develop text-mining tools to automatically extract drug interactions, adverse drug reactions (ADRs), and related events from biomedical text.
• Improve Drug Safety Monitoring: Train models to monitor and assess medication safety profiles, including patient-specific factors and potential drug-drug interactions.
• Facilitate Pharmacovigilance Research: Provide a labeled dataset for training, testing, and benchmarking novel machine-learning algorithms in pharmacovigilance.

Methodology:

Domain experts manually annotated 597 MEDLINE abstracts following detailed annotation guidelines, and inter-annotator agreement was measured (F-score 92.6% for named entities and 78.4% for complex events under relaxed matching).