In Silico Prediction of Drug Side Effects
Michael J. Keiser · Methods and principles in medicinal chemistry · 2015
Established drugs frequently lack selectivity, and a third to half of research compounds in pharmacological databases are reported to be promiscuous. Correspondingly, attempts to predict adverse drug reactions (ADRs) must take multitarget drug activity into account. Systems pharmacology approaches do so by combining systems biology with chemoinformatic inference to investigate how molecular interactions drive whole-body outcomes. These multiscale models form networks of target perturbation and biological effect that, when organized into layers of increasing complexity, comprise the drug response of a binding site, cell, tissue, organ, or entire patient in the clinic.Systems pharmacology's application to the prediction of antitarget-mediated side effects is the focus of this chapter. In broad strokes, we consider in silico methods to predict drug off-target activity from a chemocentric perspective and then turn to the analysis of off-targets as antitargets and the means by which we may associate specific side effects with them.To address this, we survey current techniques in the systematic inference of drug off-targets, such as the Similarity Ensemble Approach (SEA). These target predictions, when taken together across collections of drugs and drug-like molecules, may be organized into global pharmacological networks linking small molecules to their known and putative protein targets.Drug-to-target networks serve as a foundation for multiscale models incorporating the more complex phenotypic, disease, and ADR outcomes we ultimately seek to predict. However, the associations known between protein targets and the biological outcomes of their modulation are at times surprisingly tenuous. Whereas certain antitargets, such as the hERG channel, are well accepted, it is far from clear how one might comprehensively ask what antitarget role an arbitrary target plays, much less whether certain targets are only antitargets when perturbed together. This chapter consequently addresses the means by which off-targets might be sifted for antitargets.Looking forward, a long-standing limitation in the annotation of ADRs to new antitargets has been the inability to perform forward chemical and genetic perturbations confirming these mechanistic hypotheses in sufficiently relevant models of human patients. One approach to this, complementary to computational models, has been the emergence of rapid, whole-organism and cell-based model systems tractable to genetic perturbation that can be assayed for quantitative phenotypic readouts. The chapter thus concludes with a consideration of the current advances, challenges, and opportunities in the generation and analysis of these phenotypic data, including their use within the broader systems pharmacology context introduced at its beginning.