Adverse Drug Events: Database Construction and in Silico Prediction

Adverse
drug events (ADEs) are the harms associated with uses of given medications
at normal dosages, which are crucial for a drug to be approved in
clinical use or continue to stay on the market. Many ADEs are not
identified in trials until the drug is approved for clinical use,
which results in adverse morbidity and mortality. To date, millions
of ADEs have been reported around the world. Methods to avoid or reduce
ADEs are an important issue for drug discovery and development. Here,
we reported a comprehensive database of adverse drug events (namely
MetaADEDB), which included more than 520 000 drug–ADE
associations among 3059 unique compounds (including 1330 drugs) and
13 200 ADE items by data integration and text mining. All compounds
and ADEs were annotated with the most commonly used concepts defined
in Medical Subject Headings (MeSH). Meanwhile, a computational method,
namely the phenotypic network inference model (PNIM), was developed
for prediction of potential ADEs based on the database. The area under
the receive operating characteristic curve (AUC) is more than 0.9
by 10-fold cross validation, while the AUC value was 0.912 for an
external validation set extracted from the US-FDA Adverse Events Reporting
System, which indicated that the prediction capability of the method
was reliable. MetaADEDB is accessible free of charge at http://www.lmmd.org/online_services/metaadedb/. The database and the method provide us a useful tool to search
for known side effects or predict potential side effects for a given
drug or compound.