|
Supplemental Table 3 : Permutation analysis and false discovery rate of probe sets selected by Wilcoxon rank sum test and t-test.
Permutation analysis (N=1000) was computed for each dataset (prednisolone (PRED), vincristine (VCR), asparaginase (ASP), daunorubicin (DNR) using (a) all patients and (b) only patients with B-lineage acute lymphoblastic leukemia. For each P-value (α) using Wilcoxon rank sum test, the number of probe sets (N), the false discovery rate (FDR) and the P-value are listed. Numbers in red indicate the subset of genes chosen for further analyses (Supplemental Table 4) based on a low P-value, an FDR less than 10 percent when possible, and between 25 and 100 genes for each drug. Similar results were obtained with probe sets selected by t-test using (c) all patients and (d) only patients with B-lineage.
a. All patients
| |
α = 0.0001 |
α = 0.0005 |
α = 0.001 |
drug |
N |
FDR [%] |
P-value |
N |
FDR [%] |
P-value |
N |
FDR [%] |
P-value |
PRED |
11 |
10 |
0.021 |
32 |
18 |
0.017 |
53 |
22 |
0.019 |
VCR |
7 |
13 |
0.028 |
76 |
15 |
<0.001 |
76 |
16 |
<0.001 |
ASP |
24 |
5 |
0.007 |
91 |
6 |
<0.001 |
135 |
9 |
<0.001 |
DNR |
2 |
22 |
0.060 |
11 |
45 |
0.080 |
27 |
45 |
0.065 |
b. B-lineage
| |
α = 0.0001 |
α = 0.0005 |
α = 0.001 |
drug |
N |
FDR [%] |
P-value |
N |
FDR [%] |
P-value |
N |
FDR [%] |
P-value |
PRED |
19 |
6 |
0.007 |
57 |
10 |
0.002 |
92 |
13 |
0.003 |
VCR |
22 |
6 |
0.004 |
74 |
8 |
<0.001 |
138 |
10 |
<0.001 |
ASP |
67 |
2 |
0.002 |
202 |
3 |
<0.001 |
279 |
4 |
<0.001 |
DNR |
5 |
22 |
0.029 |
25 |
26 |
0.011 |
38 |
32 |
0.014 |
c. All patients
| |
α = 0.0001 |
α = 0.0005 |
α = 0.001 |
drug |
N |
FDR [%] |
P-value |
N |
FDR [%] |
P-value |
N |
FDR [%] |
P-value |
PRED |
10 |
12 |
0.058 |
33 |
17 |
0.021 |
49 |
24 |
0.034 |
VCR |
7 |
8 |
0.032 |
55 |
11 |
0.012 |
91 |
13 |
0.012 |
ASP |
21 |
5 |
0.028 |
84 |
7 |
<0.001 |
120 |
9 |
<0.001 |
DNR |
3 |
29 |
0.556 |
19 |
34 |
0.188 |
34 |
37 |
0.164 |
d. B-lineage
| |
α = 0.0001 |
α = 0.0005 |
α = 0.001 |
drug |
N |
FDR [%] |
P-value |
N |
FDR [%] |
P-value |
N |
FDR [%] |
P-value |
PRED |
26 |
4 |
0.014 |
69 |
8 |
0.003 |
98 |
12 |
0.003 |
VCR |
33 |
3 |
0.016 |
101 |
5 |
<0.001 |
166 |
7 |
<0.001 |
ASP |
71 |
2 |
<0.001 |
184 |
3 |
<0.001 |
269 |
4 |
<0.001 |
DNR |
8 |
12 |
0.140 |
34 |
18 |
0.043 |
50 |
26 |
0.068 |
In each random permutation, the class label (resistant or sensitive) was randomly assigned to each patient and genes were reselected using Wilcoxon rank sum test and t-test based on the random labels. The overall significance (Pα) of the model was estimated using the following formula:

Where α is the P-value using Wilcoxon rank sum test or t-test; q-valueαrandom is estimated q-value given a using random class label; q-valueαobs is the observed q-value given α using the observed class label. |