Figure 10

Figures of merit derived from affycomp. Panels A and B show figure of merit "5b" generated by affycomp for the HG-U95A spike in experiment, i.e. ROC curves for the task of identifying spiked in probe sets with known fold change of 2. Panel A is limited to 100 false positives, as suggested by Cope et al., while Panel B shows the same, but for all genes in the data set. Legend entries with caC denote that caCORRECT has been run on the data, and HL denotes that Harshlighting has been run on the data. "taxy" refers to the caCORRECT model of gene expression, while "rma", "mas5" and "plier" are computed with their respective R implementations. Panel C shows residuals (as colored in figure 8 and figure 2) for "chip 54" from the experiment, which has been hard-coded to be excluded from the analyses by the authors of affycomp. Panel D shows residuals for "chip21." caCORRECT improves performance for RMA, TAXY, and PLIER, but is unable to help the poorly-performing MAS5.0 for these data.