Predicting toxicity through computers: a changing world

Abstract The computational approaches used to predict toxicity are evolving rapidly, a process hastened on by the emergence of new ways of describing chemical information. Although this trend offers many opportunities, new regulations, such as the European Community's 'Registration, Evaluation, Authorisation and Restriction of Chemicals' (REACH), demand that models be ever more robust. In this commentary, we outline the numerous factors involved in the evolution of quantitative structure-regulatory activity relationship (QSAR) models. Such models not only require powerful tools, but must also be adapted for their intended application, such as in using suitable input values and having an output that complies with legal requirements. In addition, transparency and model reproducibility are important factors. As more models become available, it is vital that new theoretical possibilities are embraced, and efforts are combined in order to promote new flexible, modular tools.