Fig 1.
Overview of the MCF algorithm.
Fig 2.
Comparing the performance of MCF to MGE-SVM across integrated cancer-type datasets.
(A) A bar plot describing the predicted AUC obtained over the combined datasets of the same cancer type using a five-fold cross validation procedure for MGE-SVM (red bars) and MCF (blue bars) classifiers. AUC denotes the area under the curve. Error bars represent one standard deviation, and p-values are for a one-sided, paired-sample t-test for the AUC of each of the five folds. (B), (C) present the receiver operating characteristic (ROC) curves obtained in the classification of the lung and breast cancer combined datasets, respectively.
Table 1.
The target Ts metabolites that MCF selected when it choses ATP as a seed (↑ denotes increased formation from ATP in cancer and ↓ denotes decreased formation from ATP in cancer compared to noncancerous tissue counterpart, Methods).
The table shows one instance of each selected target although in some cases the same target metabolite was identified in multiple compartments (e.g. UDP in the cytosol and in the mitochondria).
Fig 3.
MCF pathway utilization predicts the survival of breast cancer patients, while canonical pathways show no such signal.
Shown in (A) and (B) are the Kaplan-Meier survival curves for patients predicted by MCF and canonical pathways respectively to have the best and worst prognosis (top and bottom 10% of patients scores, respectively; Methods). (C) A scatter plot showing the correlation between the prediction classification accuracy achieved using each individual MCF pathway in the combined breast cancer data from TCGA and GEO (where they are identified) (X-label) and the C-index obtained using each such pathway in predicting patients’ survival on the (unseen) METABRIC data. (D) The canonical pathway enrichment of the reactions participating in the MCF composite pathways predictive of survival. The dashed line represents a significance threshold of 0.05 (corrected for multiple hypotheses testing).
Table 2.
summary of the datasets utilized in this work for five cancer types.
N and C stand for number of normal and cancerous samples in the data, respectively.