ANYL 193

In principle the analysis is straightforward: representative samples are analyzed and multi-variate analytical (MVA) tools used to find differences. In practice the situation is complicated because real data contains other sources of variation, for example: drug metabolites, underlying changes in subject animals, instrument changes, carryover, etc. Careful experimental design can reduce unwanted variation, but it is often necessary to eliminate suspect variables so that the effects of interest can be observed.

To help address these issues, we have developed a program called MarkerView™ Software which analyzes data sets, provides links to the raw data to visually confirm differences, allows irrelevant variables to be excluded, etc. Using data from experiments designed to identify markers of drug toxicity we illustrate how the software can be used to find uninteresting variations and show how changes in the underlying endogenous metabolites can be identified.

Experimental Methods: Samples were analyzed using an AB/Sciex QSTAR® XL QqTOF instrument by TurboIonSpray®. Urine samples were diluted x10 using mobile phase A and 10 uL and injected on to a Symmetry C18 3.5 mm (2.1x100mm) column at a flow rate of 300 uL/min. MS data was collected from 75 to 1500 amu.

The resulting data was processed with MarkerView using a proprietary algorithm to extract LCMS peaks, followed by peak alignment, normalization and analysis by Principal Component Analysis (PCA) with or without Discriminant Analysis (PCA-DA). Use was made of MarkerView's abilities to rapidly and interactively generate peak profiles (the behaviour of variables across all samples), change between group and time based displays, exclude uninteresting variables, etc.