To provide the baseline for comparisons in this analysis, we performed a Welch's unpaired t-test [24] on both the RMA values and the log base 2 of the MAS5 signal values to determine if there were differences in expression level between the treated and untreated groups. Since the MAS5 signal is bounded by 0 on the left and has a long heavy tail to the right, the MAS5 data were log-transformed for the analysis to produce a more nearly normal distribution (Fig. 1). RMA is already a log-based measurement. Significant differences were also identified using the Significance Analysis of Microarrays (SAM) algorithm [19]. SAM was performed using MAS5 data both unfiltered and filtered by fraction Present at selected thresholds. All SAM analyses were done using unpaired data, 500 permutations and the 10-Nearest Neighbor Imputer. The "delta table" from each SAM analysis was used for comparisons between these data sets.
