Figure 1.
A bridge network for bile acid metabolism for determining bridge proteins.
The overall process of the network construction is described in (B–E). (A) Structure of a bridge network, composed of a metabolic sensor (red), a metabolic enzyme (blue) and a bridge protein (gray). Metabolic enzymes catalyze the reactions of metabolites. Metabolic sensors detect the levels of intracellular metabolites. Bridge proteins link metabolic sensors and metabolic enzymes. (B) Integration of possible interactions between sensors and enzymes using protein-protein interactions (PPI) and protein-DNA interactions (PDI). Information on sensors and enzymes was collected from published studies and databases. (C) Imposing constraints on nodes and edges of an integrated network. (D) A final reference network to identify bridge proteins. (E) Selection of bridge proteins from the reference network by their bridgeness scores.
Figure 2.
p-value distributions of components of a bridge network for bile acid metabolism.
(A) p-value distributions of (i) sensor, enzyme and bridge proteins (S + E + B), (ii) sensor and enzyme proteins (S + E) and (iii) bridge proteins (B). (B) Comparisons of those p-value distributions with background p-value distribution. The statistical significance levels of shifted p-value distributions were determined by one-sided Kolmogorov Smirnov tests.
Figure 3.
Multivariate analysis of bridge proteins from different networks.
(A) Overall process of multivariate classifications using features from bridge proteins of different networks. After sorting bridge proteins by their bridgeness (), features were extracted cumulatively from top-ranked bridge proteins (
). Samples were subsequently classified by cumulatively selected features and calculated classification accuracies (
). (B) Accuracies of classifications between normal colon and primary CRC tissues. For classifications, bridge proteins were obtained from (i) a bile acid bridge network (red), (ii) a glycolysis bridge network (yellow) and (iii) an whole protein network (purple). Classification accuracies were also calculated using randomly selected proteins (black) with 95% confidence intervals (gray) on the mean classification accuracies of repeated random selections (C) Accuracies of classifications between normal colon and polyp tissues. (D) Accuracies of classifications between polyp and primary CRC tissues.
Figure 4.
Identification of the prognostic ability of markers.
Their prognostic ability was examined using a dataset of tissue samples from patients with CRC [26]. (A) Heatmap of CRC tumor samples with subgroups classified by the expression patterns of bridge proteins: BA-m1 (blue), BA-m2 (yellow) and BA-m3 (red). Prognostic ability was assessed by Kaplan-Meier survival analyses. The BA-m2 group showed the poorest prognosis. (B) Prognostic ability of our bridge proteins. (C) Prognostic ability of the ColoPrint gene set [28], with subgroups classified as col-m1 (blue), col-m2 (yellow) and col-m3 (red). (D) Prognostic ability of the Wang et al. signature gene set [27], with subgroups classified as wang-m1 (blue), wang-m2 (yellow) and wang-m3 (red). (E) Prognostic ability of p53 mutation status, mutant and wild-type. (F) Prognostic ability of mismatch repair gene (MMR) status, deficient (dMMR) and proficient (pMMR). (G) Prognostic ability of KRAS mutation status, mutant and wild-type. (H) Prognostic ability of BRAF mutation status, mutant and wild-type.
Figure 5.
Identification of the prognostic reproducibility of markers.
Their prognostic ability was examined in an independent test data [22] by supervised classifications and thus confirmed their prognostic reproducibility. (A) Prognostic ability of our bridge proteins (B) Prognostic ability determined by the ColoPrint gene set in reference [28] (C) Prognostic ability determined by the Wang et al. gene set in reference [27].