Figure 8

Estimates of the null t-statistic and p-value quartiles vary as a function of signal intensity for the Affymetrix SpikeInSubset dataset. The dataset was processed using Bioconductor implementations of the RMA, MAS, and three-step normalization functions. The x-axes correspond to the rankit (i.e., $\frac{\text{rank of value}}{\text{total}\#\text{of values}+1}$) of the log of the product of the expression means. The y-axes correspond to the observed t-test statistics and the observed two-sided p-values for the top and bottom rows, respectively. Solid and dashed grey lines indicate the theoretical medians and quartiles, respectively. The null p-values and t-statistics were modeled as a function of a 4^thorder polynomial for rankit intensity. Black and red lines correspond to the quantile regression fits for τ = 0.5 (solid) and τ = 0.25, 0.75 (dashed). The plots suggest that the null distributions may be intensity dependent, although not to the extent as was observed in the original Golden Spike datasets. Accurate quantification of the statistical significance associated with the observed intensity dependence requires an understanding of the underlying correlation structure across expression measures and remains an open research question.