A method of generating models with which to characterize selected aspects of the metabolic phenotype of subjects without dosing a test substance to those subjects or with which to predict, without dosing, the post-dose responses of subjects where those responses are dependent on metabolic phenotype, the method comprising: obtaining pre-dose data relating to a plurality of subjects before dosing with a dosing substance; obtaining post-dose data relating to the plurality of subjects after dosing with the dosing substance; and correlating inter-subject variation in the pre-dose data with inter-subject variation in the post-dose data, and generating a pre-to-post-dose predictive model on the basis of the observed correlation. The models may be used to determine selected aspects of the metabolic phenotype of a subject or to predict, without dosing, the post-dose responses of subjects. This is achieved by analysing data relating to the un-dosed subject in relation to a model describing the correlation of pre-dose and post-dose data relating to a plurality of subjects when dosed with a particular substance which challenges the biochemical transformation or pathway of interest; and generating, according to the predetermined criteria of the model, a numerical measure or classification describing the metabolic phenotype of the un-dosed subject.