The first validation set comprised 286 samples; the prediction accuracy was 87.76%, sensitivity 87.56%, specificity 88.31%, PPV 95.31% and NPV 72.34% . The second validation set comprised 198 samples; the prediction accuracy was 88.89%, sensitivity 92.54%, specificity 81.25%, PPV 91.18% and NPV 83.87% . The third validation set is composed of 97 samples; the prediction accuracy was 97.94%, sensitivity 96.43%, specificity 100%, PPV 100% and NPV 95.35% . Figure 3 shows the specificity and sensitivity values for gene sets predictive of ERBB2 status selected by using Spearman rank correlation cutoffs between 0.34 and 0.39. For the first training set , the sum of specificity and sensitivity was constant for the examined range of Spearman rank correlation cutoffs. Therefore, we used an additional set of samples for training , which led to the highest combination of specificity and sensitivity values at a cutoff of 0.35, yielding a gene INCB18424 JAK inhibitor signature consisting of 14 annotated genes and 1 probe set Everolimus mTOR inhibitor representing an unknown sequence . The ERBB2 gene and 5 other genes are part of the 17q12-q21 amplicon and are reported to be co-amplified with the ERBB2 locus . Several of these genes are represented by a number of probe sets indicating that they readily detect their cognate transcripts in breast tumor RNA samples . The remaining 8 genes comprising the candidate ERBB2 gene signature have not previously been reported to correlate with ERBB2 expression. Because our signature comprises 14 genes and one probe set representing an unannotated gene we henceforth refer to the ERBB2 predictor as the ����14-gene ERBB2 signature����. The 14-gene ERBB2 signature separated ERBB2-positive tumors from ERBB2-negative tumors with an accuracy of 93.18%, sensitivity of 77.78%, specificity of 94.94%, PPV of 63.64% and NPV of 97.40% in the 88 training samples of the first training set . The second training set comprised 144 breast tumor profiles: the prediction accuracy was 88.89%, sensitivity 59.09%, specificity 94.26%, PPV 65.0%% and NPV 92.74% . To determine whether the predictive performance of a single probe set is sufficient to determine ERBB2 status, we used the ����203497_at����, the probe set with the highest Spearman rank correlation in the 14-gene ERBB2 signature , which we termed the ����best probe set���� for the ERBB2 predictive signature. For the first training set the predictive accuracy of the ����best probe set���� was 96.59%, sensitivity 87.5%, specificity 97.5%, PPV 77.78% and NPV 98.73% . For the second training set the predictive accuracy of the ����best probe set���� was 86.11%, sensitivity 40.91%, specificity 94.26%, PPV 56.25% and NPV 89.84% . Although predictions by using ����best probe set���� in both training sets provided similar results, the sensitivity of prediction by using the ����best probe set���� in the second training set was very low, reaching 40.91%.