AI-Empowered Multimodal Data Integration and Knowledge Discovery for Pharmacovigilance

An AI-driven framework integrated over 2.47 million heterogeneous pharmacovigilance data relationships, enhancing drug-ADE knowledge with confidence scores and evidence provenance.

• Why it matters: Heterogeneous biomedical data sources vary in structure and evidence quality, making reliable knowledge extraction challenging, which limits effective pharmacovigilance and AI applications.
• What they did: The framework combines LLM-based information extraction, ontology-driven normalization, evidence assessment, and machine learning to process data from FAERS, literature, trials, and EHRs, creating a structured knowledge resource.
• The result: Incorporating this resource improved the relevance, comprehensiveness, and quality of drug-specific ADE synthesis and increased the precision and F1 score of clinical ADE relation extraction, supporting reproducible biomedical AI.