Figure 4

Flux correlations and prediction errors for the PSEUDO method. (A, B, C) Flux predictions derived using the FBA, MOMA and PSEUDO objective functions are plotted against 320 flux measurements taken from 12 different mutants in the Tomita data set. Flux values are reported as a percentage of the glucose uptake rate in each mutant. Symbols indicate the metabolic pathway associated with each flux. The FBA, MOMA and PSEUDO methods yielded Pearson correlation coefficients of 0.86, 0.84 and 0.91 respectively. The PSEUDO coefficient was significantly higher than that of the other two methods by both Meng's Z-test (p-value = 2.4·10^-12) and bootstrap resampling (p-value < 1·10^-6). (D, E, F) Histograms depicting the prediction errors for three metabolic pathways by three methods. While prediction errors were comparable among the three methods for fluxes belonging to the glycolysis and PPP class, errors in predicting TCA cycle fluxes were lower using the PSEUDO objective. Mean prediction errors of -41%, -42%, and -17% were obtained for FBA, MOMA and PSEUDO respectively. The reduced prediction errors using the PSEUDO method were significant by bootstrap resampling (p-value < 1·10^-6).