Adverse drug events: database construction and in silico prediction

Area of Science:

• Pharmacovigilance and Drug Safety
• Computational Biology
• Bioinformatics

Background:

• Adverse drug events (ADEs) cause significant morbidity and mortality, often identified post-market.
• Current methods struggle to identify all potential ADEs before drug approval.
• Millions of ADEs are reported globally, highlighting the need for better management tools.

Purpose of the Study:

• To develop a comprehensive database of drug-ADE associations.
• To create a computational method for predicting potential ADEs.
• To provide a tool for searching known and predicting unknown drug side effects.

Main Methods:

• Integrated data from various sources and employed text mining to build the MetaADEDB database.
• Included over 520,000 drug-ADE associations for 3059 compounds and 13,200 ADEs.
• Annotated all entries with Medical Subject Headings (MeSH) terms.
• Developed the phenotypic network inference model (PNIM) for ADE prediction.

Main Results:

• The MetaADEDB database contains extensive drug-ADE associations.
• The PNIM achieved high predictive performance with an AUC > 0.9 in cross-validation.
• External validation using US-FDA data yielded an AUC of 0.912, confirming PNIM's reliability.

Conclusions:

• MetaADEDB is a valuable resource for pharmacovigilance and drug discovery.
• The PNIM offers a reliable computational approach for predicting potential ADEs.
• This integrated approach enhances drug safety assessment and management.