Figure 5

ROC and PR curves of biclustering networks using modified evaluation methodology. The poor result of the previous figure (Figure 4) should be considered with regard to two important issues. First, some of the links reported in the interactome databases may be indirect rather than the direct. Second, the available interactome databases are still incomplete, so the false positive edges that are unsupported are not necessarily false and may find experimental evidence in the future. The False Positive (FP) edges could be considered True Positive (TP) if they have evidence in the literature (gold network). For example, if the inference network includes an edge between gene1 and gene3, which does not exist in the gold network, and if these two genes connect indirectly via another intermediate gene such as gene2, we can now consider the edge between gene1 and gene3 as a true positive edge. To be entirely consistent we change a TN edge into FN every time there is an interaction via an intermediate gene.