For both ethical and scientific reasons it is always important to seek new ways to make the best use of any sacrificed animal, which will lead to higher accuracy in ensuing clinical studies. In line with this thought we, by the current paper, want to present a new way towards a holistic and robust modelling of animal data, as exemplified on the murine dextran sodium sulphate model of colitis. This has been achieved by a global property screening approach, using a new way to set up data matrices for principal component analyses, applied to the characterization of the mouse model per se, but also to pharmacological treatments with small molecules. Indeed, the two strains behaved differently after the DSS challenge. This was observed by inspection of the individual animal Methoxsalen positioning in the PCA scores plot. The PCA scores are generated by estimation of the object inter-related positioning in a scatter space comprising all collected biomarker data. The method implements each biomarker data as an individual and uniquely directed axis. In the example shown in Figure 1, 21 different biomarkers were assessed, thus generating 21 uniquely directed axes. Thereafter, the PCA least squares procedure was initiated, including the estimation of new vector dimensions adhering to minimized sum of squared distance from each data point to the generated vector, and thus the first PC is generated. In cases when the model needs more than a onedimensional structure, a second and third and etc., component can be generated using the residual data. The validation for PCA model complexity was in the present study estimated by the cross-validation procedure as described in material and methods. First, a PCA-model was established using DSS- and placebotreated and healthy control mice, respectively, with the aim to predict the data from animals receiving an additional compound treatment. This is possible for any PCA model given that the same variables, i.e. colitis biomarkers in the present study, are measured coherently. The data alignment aspect is depicted in Figure 2 and such prediction is straight-forward once a PCA model has been created. The expected advantages from this approach relates to standardization. Firstly, we provide a basic model that does not change in scale, biomarker loading correlation pattern towards the model PCs, or any other model property. Secondly, when the PCA model is used for prediction of a pharmacological compound in treated animals one can assess the cohort clustering and interpret their systems pharmacology positioning via the scores plot as compared to the healthy and DSS-treated cohort Soyosaponin-Ac distributions. However, in this step it is of outmost importance over time to include model animal objects to reveal any drift in the healthy and DSS treated animal clustering, be it for animal phenotype due to e.g. health status or response to DSS etc, or for drift in the analytical measures settings. Thirdly, over time it is possible to combine any number of treatments and any comparison combination, including reference and tool compound treatments, for drug target and pharmacological mechanism investigation. By using the PCA prediction mode for new samples, i.e. predicting treated animal property positions into the initial model including the sample data from healthy and diseased animals, a robust model that facilitates interpretation between treatments over time is provided. The resulting systems pharmacology model on the present study is illustrated in Figure 3A, comprising 2 significant principal components. Finally, the PPARa-treated mice demonstrate the highest degree of disease amelioration of the three groups.