Table 4.
Accuracy of predicted TF–GO terms associations at different E-value cutoffs using the Avg-Odds global method
| E-value cutoff | Yeast |
Human |
||||||
|---|---|---|---|---|---|---|---|---|
| Pred. | 1 TP | Rec. | Prec. | Pred. | 1 TP | Rec. | Prec. | |
| 0.01 | 5.4 | 0.56 | 0.27 | *0.61 | 22.8 | 0.75 | 0.10 | *0.60 |
| 0.05 | 8.2 | 0.58 | 0.32 | *0.49 | 31.5 | 0.75 | 0.13 | 0.44 |
| 0.1 | 10.0 | 0.60 | 0.36 | *0.44 | 36.8 | 0.75 | 0.14 | 0.40 |
| 1 | 20.2 | 0.73 | 0.49 | 0.27 | 72.3 | 1 | 0.22 | 0.33 |
| 10 | 54.0 | 0.87 | 0.66 | 0.15 | 132.3 | 1 | 0.30 | 0.25 |
| 50 | 144.6 | 0.93 | 0.78 | 0.07 | 250.3 | 1 | 0.37 | 0.17 |
The columns indicate the E-value cutoff, number of TF–GO term association predictions returned by the method at that cutoff, the probability of predicting at least one true positive (1 TP; higher is better), the recall (higher is better), and the precision (higher is better). The values for yeast are averages calculated from the 43 TFs for which target GO terms exist. The values for human are averages calculated from NFKB1, SRY, CREB1 and TP53. When one or more TFs render no predictions they are excluded from the precision average (marked with ‘*’).