Figure 2

$\overline{\mathit{\text{AUPRC}}}$distribution. Comparison of the predictive models in terms of $\overline{\mathit{\text{AUPRC}}}$ learned by CLUS-HMC and NHMC (α = 0.5 and α = 0) from the most connected subsets of genes from the (a) gasch2 and (b) cellcycle datasets annotated with labels from the GO hierarchy. The horizontal axis gives the minimum relative number (in %) of interactions a gene must have in the DIP PPI network to be included in the testing data, whereas the vertical axis gives the model performance on the testing data in terms of the $\overline{\mathit{\text{AUPRC}}}$ values. At the far right (100 on the horizontal axis), we have the performance on the most-highly connected genes from the test set. At the far left (0 on the horizontal axis), we have the performance on all genes for the testing set.