Table 2.

Average AUC for RMA, RMA-noBG, MAS 5.0, dChip, and PLIER for detection of spiked-in genes in the Affymetrix Latin Square spikein experiments, according to the value of d, which is related to the log base 2 fold change between experiments.

d	DFCM	RMA	RMA-noBG	MAS 5	dChip	PLIER
Average AUC for the HG-U95Av2 Latin Square Spike-In Data Set
1	0.732	0.715	0.721	0.063	0.572	0.062
2	0.871	0.869	0.918	0.167	0.803	0.316
3	0.936	0.935	0.935	0.484	0.886	0.629
4	0.997	0.994	0.986	0.798	0.948	0.769
5	1.000	0.999	0.999	0.916	0.980	0.853
6	1.000	1.000	1.000	0.967	0.987	0.876
7	1.000	1.000	1.000	0.981	0.999	0.876
Average AUC for the HG-U133A Latin Square Spike-In Data Set
1	0.768	0.738	0.734	0.060	0.600	0.365
2	0.858	0.831	0.812	0.307	0.709	0.752
3	0.935	0.904	0.908	0.561	0.811	0.883
4	0.934	0.964	0.964	0.837	0.913	0.951
5	0.983	0.990	0.983	0.939	0.971	0.985
6	0.999	0.998	0.996	0.968	0.989	0.994
7	0.999	0.999	0.999	0.978	0.972	0.996

Larger values of d correspond to larger fold changes among spiked-in transcripts between pairs of experiments. Eighteen spike-in probes sets were used for the HG-U95Av2 experiment, and 64 spiked-in probe sets were used for the HG-U133A to calculate true and false positives, as discussed in Methods. To calculate the AUCs, the number of false positives was set to 100 for the HG-U95Av2 experiment, and 200 for the HG-U133a experiment. Numbers in bold indicate the best value for each row.