Figure 2

Number of significant genes for titration data. The number of significant genes increases with percent MCF7 for all analysis methods (Benjamini and Hochberg FDR <0.01). The separate channel analysis makes the most discoveries at each dilution. The log-ratio method with variance shrinkage is the next most powerful. The linear mixed model which is uses a genewise variance estimator is more powerful than the log-ratio method with genewise variance estimator, but not as powerful as either of the methods using variance regularization. For these data, maanova is the least powerful.