Table 4 |
||||||||||||
|
Performance on simulated data for scenario I |
||||||||||||
| Parameters |
Rules |
q |
R |
CI95 |
O11 |
O1+ |
O+1 |
FP (%) |
TP (%) |
FN (%) |
TN (%) |
Global error |
|
|
||||||||||||
| Independence case: n = 3000, common = 0, DE1 = 1000, DE2 = 800 |
||||||||||||
| Independence: signal to noise |
0.55 |
1* |
0.98-1.02 |
0† |
0† |
0† |
0 |
0 |
0 |
3,000 (100.0) |
0 |
|
| ratio = 0.4‡ |
||||||||||||
| A: n = 3000, common = 700, DE1 = 1000, DE2 = 800 |
||||||||||||
| Case A1: signal to noise ratio = 9.6‡ |
Max |
0.01 |
2.60 |
2.50-2.72 |
619 |
975 |
730 |
4 (0.2) |
615 (87.8) |
85 (12.2) |
2,296 (99.8) |
89 |
| Double |
0.06 |
2.04 |
1.97-2.19 |
676 |
1,095 |
877 |
29 (1.3) |
647 (92.4) |
53 (7.6) |
2,271 (98.7) |
82 |
|
| Min§ = 81 |
||||||||||||
| Case A2: signal to noise ratio = 1.6‡ |
Max |
0.01 |
4.72 |
4.19-5.29 |
86 |
346 |
157 |
1 (0.0) |
85 (12.1) |
615 (87.9) |
2,299 (100.0) |
616 |
| Double |
0.08 |
2.01 |
1.90-2.20 |
212 |
677 |
459 |
28 (1.2) |
184 (26.3) |
516 (73.7) |
2,272 (98.8) |
544 |
|
| Min§ = 535 |
||||||||||||
| B: n = 3000, common = 200, DE1 = 700, DE2 = 500 |
||||||||||||
| Case B1: signal to noise ratio = 9.6‡ |
Max¶ |
0.01 |
1.72 |
1.58-1.86 |
185 |
691 |
467 |
8 (0.3) |
177 (88.5) |
23 (11.5) |
2,792 (99.7) |
31 |
| Min§ = 31 |
||||||||||||
| Case B2: signal to noise ratio = 1.6‡ |
Max |
0.01 |
2.98 |
2.38-3.71 |
36 |
250 |
145 |
3 (0.1) |
33 (16.7) |
167 (83.3) |
2,797 (99.9) |
170 |
| Double |
0.03 |
2.03 |
1.67-2.40 |
57 |
355 |
236 |
11 (0.4) |
46 (23.0) |
154 (77.1) |
2,789 (99.6) |
165 |
|
| Min§ = 165 |
||||||||||||
| C: n = 3000, common = 100, DE1 = 500, DE2 = 400 |
||||||||||||
| Case C1: signal to noise ratio = 9.6‡ |
Max¶ |
0.01 |
1.48 |
1.30-1.67 |
95 |
500 |
383 |
7 (0.2) |
88 (88.4) |
12 (11.6) |
2,893 (99.8) |
19 |
| Min§ = 19 |
||||||||||||
| Case C2: signal to noise ratio = 1.6‡ |
Max |
0.01 |
2.93 |
2.16-3.83 |
20 |
214 |
96 |
3 (0.1) |
17 (16.6) |
83 (83.4) |
2,897 (99.9) |
86 |
| Double |
0.02 |
2.16 |
1.63-2.81 |
26 |
262 |
134 |
5 (0.2) |
21 (21.0) |
79 (79.0) |
2,895 (99.8) |
84 |
|
| Min§ = 84 |
||||||||||||
|
|
||||||||||||
|
Average simulation results: we show the results from the joint model on one case of simulated data for independent experiments and six cases of simulated data for two associated experiments. The simulation scenario consists of four groups of genes: differentially expressed DE in both experiments, differentially expressed in only one experiment (DE1 and DE2 respectively), and differentially expressed in neither experiment. For the Independence case, the number of genes differentially expressed in both experiments was set to 0. We present two decision rules: the threshold associated with the maximum R(q) is qmax and the threshold associated with the R(q) ≥ 2 is q2 (called 'double' in the table). We define qmax = arg max{Median(R(q) | O, n) over the set of values of q for which CI95(q) excludes 1} and q2 = max{over the set of values of q for which CI95(q) excludes 1 and Median(R(q) | O, n) ≥ 2}. We averaged the results over 50 repeats for each case. *In case of independence it is still possible to calculate he maximum of R(q), but it is not significant, so there is no associated list of common genes. †All the CIs contain 1, so no genes are called in common; thus, there are no FP. ‡The signal to ratio is calculated as E(Ga(shape, 1/scale))/(r1/2 + r2/2). §Minimum global error (observed). ¶There is no ratio larger than 2 and only the maximum rule has been reported. |
||||||||||||
|
Blangiardo and Richardson Genome Biology 2007 8:R54 doi:10.1186/gb-2007-8-4-r54 |
||||||||||||
