The accuracy of the reconstructed context-specific GEMs.

(A) Pearson's pairwise correlation plot shows correlation of gene expression P/A calls (right-bottom; green) and model structures (left-upper; purple) between cancer types. Values next to the correlation plot represent final correlation between gene expression P/A calls and the network structure for each cancer type, values in parentheses represent P-values of significance (Figure S6 in Text S1). (B) The accuracy of the predicted flux changes from the reconstructed models. The prediction of gene expression changes from changes in flux was tested against gene expression data sets for cancer vs. normal. The black bar depicts the accuracy of the flux prediction from the reconstructed models. The grey bar denotes the median value of 10,000 accuracies from the random flux. The error bar shows the standard error mean (SEM) of 10,000 random accuracies. (C) The percent stacked column chart shows the percent of genes for three categories of gene knockout effects (Essential: knockout effect that reduced the growth rate more than 10%, 1∼10% reduced: knockout effect that reduced the growth rate from 1 to 10%, No change: knockout does not cause any change). (D) The accuracy of a model is evaluated by enrichment analysis of experimentally validated essential genes among the predicted essential genes with the model. A P-value from the hypergeometric test is converted to a z-score. Higher z-scores correspond to more significant enrichment (Orange: z-scores of essential genes, yellow: z-scores of genes causing 9% growth reduction, sky-blue: causing 5% growth reduction, 1%: causing 1% growth reduction).