Table 2 Results for the data from Golub et al (detection of differential expression between ALL and AML) and from Armstrong et al (detection of differential expression between ALL and AML, between ALL and MLL and between MLL and AML)

From: Balancing false positives and false negatives for the detection of differential expression in malignancies

 

Golub et al (ALL-AML)

Armstrong et al (ALL-AML)

Armstrong et al (ALL-MLL)

Armstrong et al (MLL-AML)

N

7129

12 582

12 582

12 582

N 0

3876

3084

8119

4527

N 1

3253

9498

4463

8055

AUC (%) (95% CI)

91.39 (90.68–92.10)

95.13 (94.78–95.48)

85.98 (85.24–86.72)

94.83 (94.46–95.20)

α opt

0.18 (=p3429)

0.11 (=p8633)

0.22 (=p5193)

0.13 (=p7589)

SENSopt (%)

84.03

87.26

76.75

86.97

SPECopt (%)

82.06

88.56

77.97

86.78

SENSopt+SPECopt (%)

166.09

175.82

154.71

173.76

  1. N=total number of genes; N0=number of genes without actual differential expression; N1=number of genes with actual differential expression; AUC=area under the ROC curve; αopt=rejection level where the optimal balance between specificity and sensitivity is reached (i.e., the rejection level that maximises the sum of sensitivity and specificity – for the first two columns, these are also the rejection levels associated with the points on the ROC curves in Figure 1 with tangent lines with slope 1); SENSopt=sensitivity at αopt; SPECopt=specificity at αopt.
Back to article page