Through examination of inter-platform correlation and expressed versus unexpressed gene properties, we have demonstrated that variance measurement may aid in the determination of whether or not a gene is expressed, particularly at low microarray expression values.
RNA species that vary together under a range of conditions are likely to be under common regulation, and indeed, sets of "co-expressed" genes generated by clustering of microarray expression values have proven useful for identifying potential regulatory elements and transcription factor binding sites [ 1- 5].
It is important to note that many genes identified as expressed by the nCounter analysis system had microarray expression values lower than those identified as unexpressed, confirming on a new platform the observation made by Irizarry and colleagues [ 35, 36] that probe-effects prevent strict, experiment-wide thresholding of expressed genes.
To get more precise error estimates, the shrinkage t-score makes therefore use of the fact that the variability of microarray expression values is governed by methodical factors which allow to express the measurement error as a function of the expression level [ 43, 51].
In order to select control genes that were well-expressed but not completely saturated, we filtered for genes with mean microarray expression values between 8 and 12 (see Additional file 2 for variance-expression relationship).
High correlations between qRT-PCR and microarray expression values were validated (Additional file 4: Table S3).
HCL was performed on the microarray expression values, where the average of all samples in a tumor category was calculated and further processed.
Here, the known concentrations of the exogenous genes can be used as a reference to evaluate the accuracy of the microarray expression values [1], [2].
The trend was not as evident for the decreased global miRNAs where there was slightly greater variation in miR-181b, 1280 and 432 qRT-PCR and microarray expression values; however miR-1826 levels coincided between assays (Figure 4A).
The results (FC, p-value and FDR) of the performed comparative analyses for all the probe sets encoding AMP genes are given in Table S2. Figure 1A F shows the individual microarray expression values for DEFB1, LEAP2, PYY, DEFA5, DEFA6 and DEFB4, respectively.
First, inter-sample variance may indicate transcript presence in genes with low microarray expression values.
