Fig 1.
SUMOFLUX workflow for targeted flux ratio analysis.
Input data are depicted in the dashed-line rectangles.
Fig 2.
Comparison of SUMOFLUX and analytic formula estimates for flux ratios in E. coli central carbon metabolism.
From left to right: a schematic representation of the flux ratio; density plot representing SUMOFLUX estimates versus the true flux ratios for in silico data; comparison of the SUMOFLUX and analytic formula estimates for the experimental data; density plot representing analytic formula estimates versus the true flux ratios for in silico data. Vertical error bars in the third panel represent [10–90%] SUMOFLUX prediction quantiles, horizontal error bars represent standard deviation provided with the analytic formula estimate. (a) Glycolysis versus PPP. (b) Pyruvate fraction from the E-D pathway. (c) PEP fraction from gluconeogenesis. (d) Pyruvate fraction from the malic enzyme flux. (e) Oxaloacetate fraction from anaplerosis from PEP. Ratios (a)-(c) were estimated from [1-13C] glucose experiment, ratios (d) and (e) were estimated from 20% [U-13C] and 80% naturally labeled glucose experiment. 6PG– 6-phosho-D-gluconate; αKG– α-ketoglutarate; AcCoA—acetyl-CoA; E-D—Entner-Doudoroff pathway; F6P –fructose-6-phosphate; Fum—fumarate; G6P –glucose-6-phosphate; Gox—glyoxylate; ICT—isocitrate; KDPG—2-Keto-3-deoxy-6-phosphogluconate; MAE—mean absolute error; Mal—malate; PCC—Pearson correlation coefficient; PEP—phosphoenolpyruvate; PGA—phosphoglycerate; PPP—pentose phosphate pathway.
Fig 3.
SUMOFLUX is robust in terms of experimental noise and exchange flux magnitude.
(a) Mean absolute errors on the testing dataset of five flux ratio predictors applied to in silico data with different amount of measurement noise and exchange flux magnitude. The dashed rectangle indicates the normal range of noise (0.01) and exchange flux magnitude (10 times the net flux). (b) Mean absolute errors on the testing datasets with different noise levels of five flux ratio predictors trained on datasets with different amount of measurement noise. The exchange flux magnitude was set to 1 for all datasets. (c) Mean absolute errors on the testing dataset with different exchange flux magnitudes of five flux ratio predictors trained on datasets with different values of exchange flux magnitude. The noise level was set to 0.01 for all datasets. E-D—Entner-Doudoroff pathway, MAE—mean absolute error; PPP—pentose phosphate pathway.
Fig 4.
SUMOFLUX resolves a novel flux ratio in central carbon metabolism of E. coli.
(a) A schematic representation of the glyoxylate shunt, TCA cycle and anaplerosis from PEP flux fractions. (b) Density plot representing SUMOFLUX estimates for the flux fraction from glyoxylate shunt versus the true flux ratios for in silico data. (c) Density plot representing SUMOFLUX estimates for the flux fraction from the TCA cycle versus the true flux ratios for in silico data. Both ratios were resolved for experiment with 20% [U-13C] and 80% naturally labeled glucose. (d) Predictions for the three flux fractions for the experimental data. αKG– α-ketoglutarate; AcCoA—acetyl-CoA; Fum—fumarate; Gox—glyoxylate; ICT—isocitrate; MAE—mean absolute error; Mal—malate; PEP—phosphoenolpyruvate; TCA—tricarboxylic acid cycle.
Fig 5.
Optimizing experimental design to improve the estimation of three flux ratios in Bacillus subtilis central carbon metabolism.
Mean absolute errors on the test dataset of three flux ratio predictors applied to in silico data simulated with different experimental setups. GC-MS—gas chromatography mass spectrometry, LC-MS—liquid chromatography mass spectrometry; LC-MS/MS—liquid chromatography-tandem mass spectrometry; LC-MRM—liquid chromatography with multiple reaction monitoring information; MAE—mean absolute error; PPP—pentose phosphate pathway.