Figure 1.
Structure of shortest paths in the yeast PPI network.
(a) Distribution of shortest-path distances in the largest connected component of the yeast PPI network; (b) an example subgraph with a hub.
Figure 2.
Distribution of DSD in the largest connected component of the yeast PPI network; the red curve represents a fitted normal distribution for comparison.
Figure 3.
An example of functional annotation with DSD.
The correct functional annotation for GLR1, on the third level of the MIPS hierarchy, 32.01.01 (oxidative stress response) is found among none of its direct neighbors, but with the node that is closest in DSD, MXR2. MXR2 is closest in DSD because it has the most similar neighborhood to GLR1.
Figure 4.
Improvement on Accuracy and F1 Score for at different neighborhood thresholds, for the majority voting algorithm in 10 runs of 2-fold cross validation for S. cerevisiae, with standard deviations.
Figure 5.
Improvement on Accuracy and F1 Score for DSD at different neighborhood thresholds, for the functional flow (FF) algorithm in 10 runs of 2-fold cross validation for S. cerevisiae, with standard deviations.
Table 1.
Summary of DSD improvements of all four methods in 2-fold cross validation (mean and standard deviation in percentage) for the PPI network of S. cerevisiae.
Figure 6.
Improvement on F1 Score for DSD using three evaluation methods: exact match, overlap depth and overlap counting, on informative GO terms for the four algorithms for S. cerevisiae in 10 runs of 2-fold cross validation.
Figure 7.
Shortest path distance and DSD distribution for S. pombe.