This resulted in a fully normalized data matrix for 485,512 CpG probes.
The slope when comparing spike-in normalized data to RPKM normalized data was effectively 1.0, indicating that the values across the transcriptome were nearly identical when normalized with the two approaches.
Extensive variation in C t values is to be expected in a non-normalized data set, and so the algorithm may not be able to effectively distinguish between stable and unstable reference genes.
The same general trends were apparent in the raw CT values for the focal genes as in the normalized data, suggesting Actin5C normalization did not skew the results.
To further eliminate genes near background in these normalized data we set a median expression threshold value of 90 for either CR or HIGHCAL.
An example of a peak not discovered in the normalized data but ranked among the top peaks unique to the analysis with experimental background for FOXA1 is shown in Figure 4.
We excluded genes that showed a standard deviation greater than 2.0 in the normalized data of both genotypes, and we started analysis with 9,640 genes and ESTs for ES cells (Table 1 (corrected)).
We excluded those genes that showed a standard deviation greater than 2.0 in the normalized data of both genotypes, therefore, we started analysis with 9,907, 12,353, and 12,359 genes and ESTs for ES cells, livers, and EFs, respectively (Table 1).
In general, the rank order differences that we observed correlate well with analog miRNA abundance, expressed as DCt in the median normalized data or DDCt in the sno135/miR-25 normalized (Statminer) analyses.
Each gene in the normalized data set was then subjected to a logistic regression analysis and the area under the ROC curve (AUC) determined.
