Identification of structure-activity relationships for adverse effects of pharmaceuticals in humans: Part B. Use of (Q)SAR systems for early detection of drug-induced hepatobiliary and urinary tract toxicities Edwin J. Matthews a, * , Carling J. Ursem a,b , Naomi L. Kruhlak a , R. Daniel Benz a , David Aragonés Sabaté c , Chihae Yang d , Gilles Klopman e , Joseph F. Contrera f a US Food and Drug Administration, Center for Drug Evaluation and Research, Office of Pharmaceutical Science, Informatics and Computational Safety Analysis Staff (ICSAS), 10903 New Hampshire Ave., Silver Spring, MD 20993, USA b GlobalNet Services, 11820 Parklawn Drive, Rockville, MD 20852, USA c Prous Institute for Biomedical Research, S.A., Provenza 388, 08025 Barcelona, Spain d Leadscope, Inc., 1393 Dublin Road, Columbus, OH 43215, USA e Multicase, Inc., Ste 305, 23811 Chagrin Blvd., Beachwood, OH 44122, USA f Computational Toxicology Services LLC, P.O. Box 1565, Olney, MD 20830, USA article info Article history: Received 9 July 2008 Available online 30 January 2009 Keywords: AERS Computational toxicology Drug adverse effects Drug-induced liver injury Drug-induced renal toxicity In silico Post-market reporting Quantitative structure-activity relationships QSAR software SAR SRS abstract This report describes the development of quantitative structure-activity relationship (QSAR) models for predicting rare drug-induced liver and urinary tract injury in humans based upon a database of post-mar- keting adverse effects (AEs) linked to 1600 chemical structures. The models are based upon estimated population exposure using AE proportional reporting ratios. Models were constructed for 5 types of liver injury (liver enzyme disorders, cytotoxic injury, cholestasis and jaundice, bile duct disorders, gall bladder disorders) and 6 types of urinary tract injury (acute renal disorders, nephropathies, bladder disorders, kidney function tests, blood in urine, urolithiases). Identical training data sets were configured for 4 QSAR programs (MC4PC, MDL-QSAR, BioEpisteme, and Predictive Data Miner). Model performance was opti- mized and was shown to be affected by the AE scoring method and the ratio of the number of active to inactive drugs. The best QSAR models exhibited an overall average 92.4% coverage, 86.5% specificity and 39.3% sensitivity. The 4 QSAR programs were demonstrated to be complementary and enhanced per- formance was obtained by combining predictions from 2 programs (average 78.4% specificity, 56.2% sen- sitivity). Consensus predictions resulted in better performance as judged by both internal and external validation experiments. Published by Elsevier Inc. 1. Introduction This is the second portion of a three-part investigation con- ducted by the US FDA’s Center for Drug Evaluation and Research (CDER), Informatics and Computational Safety Analysis Staff (IC- SAS). ICSAS is an applied regulatory research group that develops databases of toxicological and adverse human clinical information for use in data mining and quantitative structure-activity relation- ship (QSAR) modeling (Benz, 2007). In the first report we describe the creation of a human health effects database containing adverse event reporting data from two pharmaceutical post-market sur- veillance databases maintained by the FDA, the Spontaneous Reporting System (SRS) and the Adverse Event Reporting System (AERS), and from the published literature (Ursem et al., 2009). In addition we described the method that was used to identify a sub- set of pharmaceuticals that had significant hepatobiliary and uri- nary tract adverse effects (AEs). In the current report we describe the creation of QSAR models to predict hepatobiliary and urinary tract AEs of drugs based upon the molecular structure of the phar- maceutical. We employed four state-of-the-art global QSAR soft- ware programs and report the experimental parameters and methods that were needed to optimize the predictive performance of these models. In the third report we describe specific properties of the drugs that had significant hepatobiliary and urinary tract AEs. These properties include both the clinical indication(s) for which the drug was approved, and the multiple pharmacological activities of the drug predicted by QSAR programs (Matthews et al., 2009). The overall goal of that investigation was to devise a generalized in silico methodology which could be utilized to pre- dict AEs of pharmaceuticals and provide some insight into possible mechanisms of action (MOAs) responsible for the AEs. 0273-2300/$ - see front matter Published by Elsevier Inc. doi:10.1016/j.yrtph.2009.01.009 * Corresponding author. Fax: +1 301 796 9998. E-mail address: edwin.matthews@fda.hhs.gov (E.J. Matthews). Regulatory Toxicology and Pharmacology 54 (2009) 23–42 Contents lists available at ScienceDirect Regulatory Toxicology and Pharmacology journal homepage: www.elsevier.com/locate/yrtph