Integrated analysis of independent gene expression microarray datasets improves the predictability of breast cancer outcome

Table 1 Predictability of expression profiles

Training dataset	Tested on Rosetta		Tested on Stanford		Size-weighted average
	Accuracy	AUC	Accuracy	AUC	Accuracy	AUC
Individual	71.43%	0.775	70.00%	0.764	70.59%	0.767
Combined	71.43%	0.786	71.43%	0.799	71.11%	0.789

At each permutation re-sampling, SEP of testing patients was calculated with top-ranked genes obtained from training patients; the mean of accuracy and AUC was weighted by size of testing subgroups; values presented were the medians summarized from10,000 re-samplings.

ISSN: 1471-2164