Drug-induced liver injury (DILI) is one of the most common drug-induced adverse events (AE) leading to life-threatening conditions such as acute liver failure (ALF). It is also the second most common cause for post-market withdrawals or warnings. Efforts to develop new predictive methods to assess the likelihood of a drug being a hepatotoxicant have been challenging due to the complexity and idiosyncrasy of clinical manifestations of DILI. The FDA adverse event reporting system (AERS) contains post-market data that depict the morbidity of AEs. Here, a set of 2,055 unique and modelable drug entities with 13,689 drug-AE combinations was extracted from the AERS database using 38 hepatotoxicity-related query preferred terms. Using a calibration set containing 178 known hepatotoxicants, a decision tree based on disproportionality scores was developed and optimized to classify training set drugs as either active or inactive for hepatotoxicity. Classification accuracies were compared between Multi-Gamma Poisson Shrinker (MGPS), stratified proportional reporting ratio (PRR), and previously employed crude PRR disproportionality scoring schemes, and the resulting datasets were used to build QSAR models of DILI with commercially available software. Structural analyses of the entire training dataset yielded 297 clusters (222 singletons) versus the 250 clusters (180 singletons) obtained previously, indicating enhanced chemical space sampling. In addition, a jury-based consensus predictor composed of QSAR and expert-rule-based predictors was constructed to yield enhanced predictive power. Taken together our efforts provide a proof-of-principle for the development of predictive models for other AE endpoints.

Development of QSAR models for prediction of hepatotoxicity using post-market data from FDA's AERS database

245th National Meeting (2013)

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.