Table 4. Null subnetwork analysis.
| No. nodes | No. null subnetworks | Expected false positivesubnets per dataset | Min no. subnetsper dataset | Max FDR |
| 3 | 40124 | 4.0 | 4 | 1 |
| 4 | 3392 | 0.34 | 1 | 0.34 |
| 5 | 365 | 0.037 | 1 | 0.037 |
| 6 | 48 | 0.0048 | 1 | 0.0048 |
| 7 | 4 | 4×10−4 | 1 | 4×10−4 |
| 8 | 1 | 1×10−4 | 1 | 1×10−4 |
104 iterations of null subnetworks were generated, by the same subnetwork identification method as used for the real data sets, but based on p-values randomly sampled from a uniform distribution. The table shows the number of subnetworks of each size which were declared as significant by the subnetwork identification method based on these null p-values, out of the 104 iterations. The table also shows the minimum number of subnetworks of each size detected in any of the real data sets (see table 3), and corresponding conservative estimates of the FDR, defined as the number of false positives divided by the number of discoveries, for each size of subnetwork.