Trade-off in PLC subnetworks between performance in GO term recovery and partitioning guide genes into expected communities. (A) Example showing normalized Chi2 statistic and p-value calculations comparing guide gene distribution into PLC communities (numbers in deep blue within polygons) to the expect partitioning (left; 3 subpathways). Two PLCs (one with a good partitioning (center); one with a weak partitioning (right)) are shown here but the contingency matrix used in Chi2 calculations is described for only one of them (center). (B) Pair plot showing correlations (Spearman’s rho, asterisks show significance p < 0.001, upper panel) and scatterplots (lower panel) between average network node degree, NV AUROC, normalized Chi2, modularity and the number of guide genes in the network. Each point in the lower panels (scatterplots) represent one network for which 2 characteristics (eg NV AUROC and modularity) are compared. Data are presented for each pathway separately with a specific color. (C) The expected partitioning of phenylpropanoid related guide genes was compared to two PLC: (D) higher predictability and lower modularity (microarrays raw PCC) and (E) lower predictability and higher modularity (microarrays PCC-HRR). In (D and E), colored vertices correspond to genes encoding enzymes catalyzing steps of similar color in (C). Community (surrounded by grey polygons) numbers in (E) are indicated in deep blue and can be used to access Supplementary Table 3.