Deciphering Molecular Mechanisms of Adverse Reactions of Drugs /

Some drugs result in severe side effects. Here we explore two different routes to attack this problem: one from structural chemistry and human proteome and another from the improved statistical analysis of patient reports. 1. Therapeutic molecules interact with multiple protein targets: some modulation events due to this poly- pharmacology are desirable for therapeutic effects and some cause serious adverse reactions. The identification of drug poly-pharmacological targets in advance remains one of the major challenges in drug development. Here, we developed a drug target profiling system using a hybrid chemoinformatic and pharmacophoric approach. The two approaches complement each other since the first one is strongly dependent on the known chemical structures of the compounds while the approach based on three dimensional pharmacophoric field matching depends on the previous knowledge to a lesser degree. The profiling system not only proved to be efficient for retrospective prediction but also enables the discovery of previously unknown drug targets that may lead to side effects. 2. Independently, unexpected adverse reactions are revealed in clinical trials and, in some unfortunate cases, only in post- marketing surveillance by the FDA many years later. The FDA post-marketing data, however, have not been properly analyzed and used for quantitative characterization of the risks of severe side effects. Here we develop a profiling system--the Early Drug Alerts, EDA -- in which we separated the drug-dependent rates of adverse effects from the disease-dependent rates. The EDA also improved signal- to-noise ratio by aggregating related adverse reactions in order to overcome the issues of data fragmentation and data scarcity. Many of the established risk signals were confirmed by clinical studies while some safety alerts revealed by the EDA analysis need immediate attention. 3. Finally, by integrating off-target profile, adverse reaction profile, and gene-phenotype relationship, we discovered a reason why raloxifene, an estrogen receptor modulator, may cause thromboembolism and confirmed its binding to thrombin. This emerging profiling pipeline could be applied to raise alerts about adverse reactions in new drug candidates, discover novel anti-targets explaining side effects, repurpose drugs for new indications, and expedite the rational design of more effective, but less toxic dru