Algorithm-driven Artifacts in median polish summarization of Microarray data

BMC Bioinformatics

Table 2 comparison between RMA and tRMA in the affycomp benchmark

	MAS5	RMA	tRMA	Best possible
Signal detect slope	0.71	0.63	0.63	1
Signal detect R2	0.86	0.80	0.80	1
Obs-intended-fc slope	0.69	0.61	0.61	1
Obs-(low)int-fc slope	0.65	0.36	0.36	1
null log-fc IQR	0.85	0.19	0.20	0
null log-fc 99.9%	4.48	0.57	0.58	0
low AUC	0.07	0.40	0.39	1
med AUC	0.00	0.87	0.86	1
high AUC	0.00	0.46	0.44	1
weighted avg AUC	0.05	0.52	0.51	1
Median SD	0.63	0.11	0.12	0
low.slope	0.72	0.35	0.35	1
med.slope	0.80	0.76	0.76	1
high.slope	0.45	0.47	0.47	1

affycompII most indicative results (as in [6]) for MAS5, RMA and tRMA, spike-in HGU95 dataset. Differences between RMA and tRMA are trivial, especially when compared to other methods (see also [6]).

ISSN: 1471-2105