In vivo predictive toxicogenomics.

Reference databases consisting of large sample numbers and high-dimensional microarray data are now available for the investigation of adverse events in animal model systems such as the rat. This large volume of data, accompanied by appropriate study designs, compound and dose selection procedure, and minimization of technical and biological confounding effects, can yield successful predictive models for a variety of hypotheses. The process of training, validating, and implementing predictive models is cyclical and complex. This chapter highlights individual decisions that need to be made before, during, and after a model or set of models has been trained, with an emphasis on proper statistical methods and suitable interpretation of the results.