In silico prediction of drug safety: despite progress there is abundant room for improvement.

Predictive models for drug safety are crucial for helping to avoid costly late-stage failures. We review recent work on models for genotoxicity, liver toxicity, CYP450 inhibition and cardiotoxicity. These models have improved somewhat in recent years, and research has expanded into new frontiers, such as the prediction of liver toxicity. However, much more needs to be done.: