Figure 3. Prediction of CCPPs does not require the availability of data in all three networks.

In the figure, we examine the set of 7510 correctly predicted interactions before the 50th false positive for an SVM trained using all three diffusion kernels. The figure plots the percentage of correctly predicted protein pairs that are directly linked (“length = 1”) or linked by a path of any length (length≥1) in three, two, one or none of the three interaction networks.