Based on a devised metric for “bridgeness”, we computationally selected bridge proteins from a reference network and examined their prognostic potential in CRC. We also tested whether differences in their discriminative expression patterns in normal colon and CRC made them relevant to CRC pathogenesis. The findings indicate that bridge proteins involved in the regulation of BA metabolism may be reliable prognostic markers for CRC patients. We also assessed the prognostic reproducibility of bridge proteins through a supervised classification system. In this classification system, the previous Pazopanib dataset were used as a training data and the Sheffer et al. dataset were used as an independent test data during supervised classifications. Total 182 tumor samples of patients from the Sheffer et al. data were used, after excluding samples that were not used in the original study. This dataset contains information about gene expression and CRC-specific survival event and time. Performing supervised classification, we first determined a patient group with the poorest prognosis from the training data, after clustering patients by a hierarchical clustering and comparing survival probabilities among patient clusters. Referencing mean expression levels of patients in the poorest prognosis group as a criterion, we classified patients of the test data into poor prognoses if their correlations of gene expressions are higher than a threshold, like existing study. We calculated the correlations based on Pearson’s correlation coefficients. A threshold of a correlation coefficient deciding prognosis was obtained through cross-validated procedures using the training data. In this data set, we performed supervised classifications through five-fold crossvalidations with various thresholds and selected the best threshold that can distinguish patients into a good or poor prognosis group with the most statistical significance. The statistical significance was measured by the Kaplan-Meier method with the log-rank test. We repeated cross-validations 100 times and averaged best thresholds in all repeats as a final threshold to use. Based on the final threshold, at last, we classified patients in an independent test data with learning a training data. We performed supervised classifications by other expression signatures in a similar way. All the statistical analyses, including Kaplan-Meier survival analysis, were performed by R packages. In non-small cell lung cancer, clinicopathological staging according to the TNM classification is still the main delimiter to classify patients with a distinct outcome. Unfortunately, of the patients diagnosed with early stage disease almost 30% to 40% will present tumor recurrence within two years after surgical resection. Since it has been shown that adjuvant chemotherapy can improve the survival of patients with resected stage II-IIIa NSCLC, identification of early stage patients with poor survival is clinically relevant. Galectins are a protein family of which the members are defined by the presence of a conserved carbohydrate recognition domain. Thus far, fifteen galectins have been identified, eleven of which are also expressed in different human cells and tissues.