Table 2.
Edge-level inference results across all settings. The TPR and FDR are calculated separately for each of the two subnetworks (
and 
). The means (standard deviations) of TPR and FDR are summarized based on 100 repeated simulations. TPR is determined by the proportion of edges in 
that can be recovered by 
, and FDR is the proportion of edges in 
are not in 
. TPR = 1 and FDR = 0 suggest a perfect recovery of 
by 
. SICERS outperforms the comparable methods because the objective function can maximize the signal while suppressing noise, and thereby better recovers the underlying true 
.
|
|
|||||||
|---|---|---|---|---|---|---|---|---|
Cohen’s 
|
1.2 | 0.8 | 0.5 | 1.2 | 0.8 | 0.5 | ||
| SICERS | TPR |
|
1(0) | 0.87(0.2) | 0.91(0.19) | 1(0) | 0.9(0.2) | 0.88(0.2) |
|
1(0) | 1(0) | 1(0) | 1(0) | 1(0) | 1(0) | ||
| FDR |
|
0(0) | 0(0) | 0(0.01) | 0(0) | 0.02(0.04) | 0.02(0.04) | |
|
0(0) | 0.03(0.04) | 0.09(0.21) | 0(0) | 0.04(0.05) | 0.09(0.19) | ||
| Louvain | TPR |
|
1(0) | 1(0) | 1(0) | 1(0) | 1(0) | 1(0) |
|
1(0) | 1(0) | 1(0) | 1(0) | 1(0) | 1(0) | ||
| FDR |
|
0.21(0.1) | 0.58(0.12) | 0.44(0.11) | 0.25(0.11) | 0.58(0.12) | 0.41(0.16) | |
|
0.16(0.06) | 0.03(0.03) | 0.04(0.04) | 0.16(0.05) | 0.02(0.03) | 0.03(0.03) | ||
| Dense | TPR |
|
1(0) | 1(0) | 1(0) | 1(0) | 1(0) | 1(0) |
|
1(0) | 1(0) | 1(0) | 1(0) | 1(0) | 1(0) | ||
| FDR |
|
0(0) | 0(0) | 0(0) | 0(0) | 0(0) | 0(0) | |
|
0(0) | 0(0) | 0.35(0.27) | 0(0) | 0(0) | 0.52(0.13) | ||
| NBS | TPR |
|
1(0) | NA | NA | 1(0) | NA | NA |
|
1(0) | NA | NA | 1(0) | NA | NA | ||
| FDR |
|
0.28(0.05) | NA | NA | 0.21(0.1) | NA | NA | |
|
0.59(0.16) | NA | NA | 0.54(0.21) | NA | NA | ||
| BH-FDR | TPR | 1(0) | 0.95(0) | 0.94(0) | 1(0) | 0.94(0.01) | 0.75(0.01) | |
| FDR | 0.18(0.01) | 0.5(0) | 0.54(0) | 0.18(0.01) | 0.5(0) | 0.54(0.01) | ||