Prediction results for the superPC and COXEN methods in all breast cancer cell lines evaluated by AUC scores.

**: P<0.05,
*: P<0.1.
The AUC values are grouped by ER status: All (cells of both ER status), ER+ (ER− positive cells), and ER– (ER-negative cells) and are separated based on the cell line expression database used to create the cell line MGPs. Note that these five validation datasets (except Tabchy-TFAC and Iwamoto) were independent for the superPC prediction method, because this predictor was not pre-optimized or optimized using any of these data sets. For the COXEN prediction method, MAQC-training and Tabchy-FEC datasets were used for optimization, and therefore the remaining three datasets were truly independent validation sets for this method.