Source | Target | Method | Connectivity | Intermediate | Precision | Recall |
---|
Ste3 | Ste12 | IFDSD | Yes | Yes | 0.67 | 0.71 |
 |  | ILP | Depend on λ | Yes | 0.56 | 0.36 |
 |  | EC | Depend on cutoff | No | 0.48 | 0.79 |
Ras2 | Ste12 | IFDSD | Yes | Yes | 0.17 | 0.63 |
 |  | ILP | Depend on λ | Yes | 0.16 | 0.38 |
 |  | EC | Depend on cutoff | No | 0.16 | 0.75 |
Mid2 | Rlm1 | IFDSD | Yes | Yes | 0.27 | 0.57 |
 |  | ILP | Depend on λ | Yes | 0.18 | 0.57 |
 |  | EC | Depend on cutoff | No | 0.29 | 0.71 |
Sln1 | Hog1 | IFDSD | Yes | Yes | 0.60 | 1.00 |
 |  | ILP | Depend on λ | Yes | 0.33 | 1.00 |
 |  | EC | Depend on cutoff | No | 0.86 | 1.00 |
- Four merits were compared among IFDSD, ILP and EC based on the yeast MAPK pathways. "Connectivity" and "intermediate" were about edges while precision and recall were about the nodes. Pathways predicted by IFDSD are always connected but the connectivity of the pathways predicted by ILP and EC depends on the parameters because they could filter the less-weighted edges. Since the nodes except the source and the target should transfer information from the source to the target, these "intermediate" nodes should have more than two edges linked to them. IFDSD and ILP always generate pathways satisfying this request whereas EC can not. Making sure the connectivity of the predicted pathways, the precision and recall were calculated by selecting the optimal parameters for each method on the yeast MAPK pathways.