The MA plot shows log fold change as a function of mean log expression level. A set of 14 arrays representing a single experiment from the Affymetrix spike-in data are used for this plot. A total of 13 sets of fold changes are generated by comparing the first array in the set to each of the others. Spiked-in genes are symbolized by numbers representing the nominal $\log_2$ fold change for the gene. Non-differentially expressed genes with observed fold changes larger than 2 are plotted in red. All other probesets are represented with black dots.

The MA plot shows log fold change as a function of mean log expression level. A set of 28 arrays representing a single experiment from the Affymetrix spike-in data are used for this plot. Fold changes are generated for all possible comparisons of the the first 14 arrays and the second 14 arrays. Spiked-in genes are symbolized by numbers representing the nominal $\log_2$ fold change for the gene. Of the genes that are spiked to be differentially expressed, only genes with small nominal fold changes are shown. The colors represent four different groups: nominal concentration of genes being compared less than or equal to 2 picoMolar (blue), between 4 and 32 picoMolar (green), greater than or equal to 64 picoMolar (blue). Non-differentially expressed genes with observed fold changes larger than 2 are plotted in red. All other probesets are represented with black dots.

For each gene, and each experimental condition in the dilution data set, we calculate the mean log expression and the observed standard deviation across 5 replicates. The resulting scatterplot is smoothed to generate a single curve representing mean standard deviation as a function of mean log expression.

For each non-spiked-in gene in the 28 arrays used in Figure 1b, we calculate the mean log expression and the observed standard deviation across the 28 replicates. The resulting scatterplot is smoothed to generate a single curve representing mean standard deviation as a function of mean log expression.

This plot, using the GeneLogic dilution data, shows the sensitivity of fold change calculations to total RNA abundance. Average log fold-changes between liver and CNS for the lowest concentration and the highest in the dilution data set are computed. Orange and red color is used to denote genes with $M_{6g}-M_{1g}$ bigger than $\log_2(2)$ and $\log_2(3)$ respectively. The rest are denoted with black.

Average observed $log_2$ intensity plotted against nominal $log_2$ concentration for each spiked-in gene for all arrays in Affymetrix spike-In experiment. The dashed line has the ideal slope of 1.

For the GeneLogic dilution data, log expression values are regressed against their log nominal concentration. The slope estimates are plotted against average log intensity across all concentrations.

Using the 28 arrays of Figure 1b, we compute local slopes. As the slopes shown in Figure 4a, the local slopes represent the expected observed log fold-change for probesets with true fold-change of 2 but they are presented as a function of the total nominal probeset concentration in the two samples being compared. In theory the local slopes should be one so we show the bias (difference between the observed local slope and one).

A typical identification rule for differential expression filters genes with fold change exceeding a given threshold. This figure shows average ROC curves which offer a graphical representation of both specificity and sensitivity for such a detection rule. Average ROC curves based on comparisons with nominal fold changes ranging from 2 to 4096.

As 5a, but with nominal fold changes equal to 2.

As 5a, but for comparisons with both nominal concentrations less than or equal to 2 picoMolar.

As 5a, but for comparisons with both nominal concentrations between 4 and 32 picoMolar and nominal fold changes less or equal to 4.

As 5a, but for comparisons with both nominal concentrations larger or equal to 64 and with nominal fold changes less than or equal to 4.

Observed log fold changes plotted against nominal log fold changes. The dashed lines represent highest, 25th highest, 100th highest, 25th percentile, 75th percentile, smallest 100th, smallest 25th, and smallest log fold change for the genes that were not differentially expressed.

Like 6a, but the observed fold changes were calculated for spiked in genes with nominal concentrations no higher than 2pM.