A Method to Calibrate Metabolic Network Models with Experimental Datasets

A method to calibrate stoichiometric coefficients values related to uncharacterized or lumped reactions of metabolic network models is presented. The method finds coefficients values that produce a model version that best fits multivariable experimental data. The method was tested with a metabolic network of 44 metabolites and 49 stoichiometric reactions, with four reactions having undetermined stoichiometric coefficients values. A total of 1320 model versions with different combinations of stoichiometric coefficient values were generated. Experimental data was used to produce a calibration curve and different fitness scores were used to evaluate the accuracy of flux balance analysis (FBA) simulations of these model versions to reproduce the experimental data. The model version with highest fitness to the experimental data was found using Mean Relative Error (MRE) scores and auto-scaled transformation of estimated datasets.