Improving metabolic flux predictions using absolute gene expression data

Abstract

Background: Constraint-based analysis of genome-scale metabolic models typically relies upon maximisation of a cellular objective function such as the rate or efficiency of biomass production. Whilst this assumption may be valid in the case of microorganisms growing under certain conditions, it is likely invalid in general, and especially for multicellular organisms, where cellular objectives differ greatly both between and within cell types. Moreover, for the purposes of biotechnological applications, it is normally the flux to a specific metabolite or product that is of interest rather than the rate of production of biomass per se.

Results: An alternative objective function is presented, that is based upon maximising the correlation between experimentally measured absolute gene expression data and predicted internal reaction fluxes. Using quantitative transcriptomics data acquired from Saccharomyces cerevisiae cultures under two growth conditions, the method outperforms traditional approaches for predicting experimentally measured exometabolic flux that are reliant upon maximisation of the rate of biomass production.

Conclusion: Due to its improved prediction of experimentally measured metabolic fluxes, and of its lack of a requirement for knowledge of the biomass composition of the organism under the conditions of interest, the approach is likely to be of rather general utility. The method has been shown to predict fluxes reliably in single cellular systems. Subsequent work will investigate the method's ability to generate condition- and tissue-specific flux predictions in multicellular organisms.

Figure 1

Steps undertaken in constraining metabolic models with gene expression data. The approach is applicable to genome-scale metabolic model that contain gene-protein-reaction (GPR) relationships. Absolute gene-expression data is mapped to individual reactions following the Boolean logic described in the “Mapping gene expression data to metabolic reactions” section of the Methods. Correlation between this gene-expression data and metabolic fluxes is maximised by following a three step algorithm comprising of: i) maximising the correlation between the initial set of irreversible reactions and the experimental data; ii) performing flux variability to determine additional reactions that must now be unidirectional; iii) repeating this cycle of maximising correlation until no extra irreversible reactions are found through flux variability analysis. The solution predicts exometabolic fluxes that can then be compared to those generated experimentally.