Figure 1 shows the specificity and sensitivity values for sets of genes predictive of ER status selected by using Spearman rank correlation cutoffs between 0.42 and 0.48. To find the most predictive set of genes, we selected those that yielded the highest combination of specificity and sensitivity values. The identified gene signature consisted of 35 probe sets, representing 24 annotated genes . Of these 24 genes, one is the ESR1 itself, whereas 11 are related to the expression of the ER: the latter include genes whose expression correlates positively with that of the ER ; genes whose expression is positively regulated by the ER ; and a gene located in close proximity to ESR1 , and whose expression is therefore positively PD-0325901 supplier correlated with that of the ER. Importantly, several of these genes are represented by multiple probe sets indicating that they PF-04217903 c-Met inhibitor robustly detect their cognate transcripts in breast tumor RNA samples . Twelve remaining genes have not been previously associated with ER status. Interestingly, SCUBE2 is reported to positively correlate with PR status . Because our ER signature comprises 24 genes and one probe set for an unknown gene, we refer to the signature as the ����24-gene ER signature����. The 24-gene ER signature separated ER-positive tumors from ER-negative tumors with an accuracy of 88.66%, sensitivity of 91.18%, specificity of 88.26%, PPV of 98.43% and NPV of 55.36% in the 247 training samples . To determine whether the predictive performance of a single probe set is sufficient to determine ER status of a sample we used ����205225_at����, the probe set with the highest Spearman rank correlation in the 24-gene ER signature , which we termed ����best probe set���� for the ER predictive signature. It is of interest, that the ����best probe set���� was the same probe set conventionally used to determine ER status . The prediction accuracy of the ����best probe set���� was 89.07%, sensitivity 89.67%, specificity 85.29%, PPV 97.45% and NPV 56.86% . Both the sensitivity and specificity of prediction by using the ����best probe set���� were lower than were the sensitivity and the specificity of the prediction using the 24-gene ER signature, indicating that the predictive performance of the single ����best probe set���� is not as high as the performance of our signature. We subsequently tested the predictive performance of the 24- gene signature in 5 independent validation datasets .