A quick method of checking the reproducibility of ICs is to plot the R0 values as a line graph, which given equal amounts of high quality RNA across all samples, should give a straight line. This is of course an ideal, and due to differences in RNA quality, spectrophotometric inaccuracy, differences in RT efficiency and biological variance, such a plot is likely to be far from linear. However, a second IC conducted on the same cDNA should provide a similar profile if IC expression accurately reflects RNA loading, and indeed this is the case (Figure 6.4A). As the level of IC expression may differ quite dramatically, samples should be normalized to the mean to transform the data to a common scale. Additional ICs should provide comparable profiles, with deviations due to differential regulation, differences in expression variance and measurement errors. The presence of a common pattern of expression with different ICs should be representative of RNA loading, and differences in the expression profile of ICs may indicate that one of the ICs is affected by the experimental procedure. Such an approach provides additional confidence that the normalization procedure used is in fact reliable.
A quick method of checking the suitability of ICs is to calculate the mean squared error (MSE) of the expression for each sample (mean normalized as noted above). This is simply a measure of the within sample variance, which will be low if the expression patterns of all IC genes are comparable across all samples, and high if the expression profiles are dissimilar. The contribution of each IC and each sample to the total sum of squares (SStota¡) can be used to determine which samples and/or genes are the greatest source of variance, and may be applied as a basis for sample or IC exclusion (Figure 6.4B and 6.4C, respectively).
Was this article helpful?