Figure 3: Lipidomics dataset: networks.

(a) The PC-corr network was constructed according to the loadings of PC2 (Fig. 2c), since PCA could significantly (p < 0.001) separate F-noCC from M, therefore reflects discriminative network modules related to real (absence of contraceptive effect) gender difference. Red nodes indicate higher lipid abundance in F-noCC, while black nodes indicate higher lipid abundance in M. (b) The P-value network is a correlation network (Pearson-correlation cut-off = 0.6) of the lipids that are significantly different (Mann-Whitney test, Benjamin-adjusted p-value < 0.05) between F-noCC and M. Again, the node colour has the same meaning of plot in (a). Modules circled with a dashed line are present in both networks (a,b), while modules circled with a black solid line are present exclusively in one of the two networks. (c) The P-value MI network is a mutual information network (mutual-information cut-off = 0.6) of the lipids that are significantly different (Mann-Whitney test, Benjamin-adjusted p-value < 0.05) between F-noCC and M, similarly to the P-value network. Again, the node colour has the same meaning of plot in (a,b). In the P-value MI network, all the edges between lipids were already inferred by the P-value network, that contains even more lipid interactions. (d) For F-noCC vs M, the PC-corr network - constructed according to the PC2 significant discrimination of (Fig. 2c) - and the P-value and P-value MI networks (cut-off = 0.6) show differences in the numbers of nodes and edges, as illustrated in the proportional Venn diagrams. Some lipids are present only in the PC-corr network (as shown in the box on the left), while much less are present exclusively in the P-value network and, for some of them, also in the P-value MI network (as shown in the boxes on the right).