Metabolism is central to cell physiology and metabolic disturbances play a role in numerous disease states. Despite its importance, the ability to study metabolism at a global scale using genomic technologies is limited. In principle, complete genome sequences describe the range of metabolic reactions that are possible for an organism, but cannot quantitatively describe the behaviour of these reactions. Gene expression data provide global insight into the regulation of metabolic reactions, but methods for inferring the behaviour of metabolic networks, and particularly metabolic flux, from these data are limited. There is thus a need to develop computational approaches that utilize available genomic data to make inferences about metabolism at the level of large scale metabolic networks.
One approach to computationally studying metabolism is to develop detailed models based on coupled differential equations describing the dynamics of enzyme action. Such models, however, require measuring numerous kinetic parameters that can be prohibitively difficult for large systems and for organisms – such as infectious disease agents – that are difficult to work with experimentally.
Flux balance analysis (FBA) is an alternative approach to modeling metabolism without developing detailed simulation models that include enzyme kinetics 
. It exploits the fact that the stoichiometries of metabolic reactions are not organism-dependent but are fixed by chemistry and mass balance. Moreover, the availability of complete genome sequences is enabling the reconstruction of metabolic networks whose constituent reactions have known stoichiometries. Flux balance analysis also exploits the fact that enzyme dynamics occur quickly compared, for example, to regulatory changes in gene expression: when the relevant laboratory time period (often hours) is much longer than the chemical reaction times (typically minutes), transient dynamics last for only a small portion of time period considered, after which the metabolic network functions at steady state. FBA is a method for utilizing universal reaction stoichiometries to predict a network's capability to produce a metabolic objective under steady-state conditions.
Briefly, FBA represents a metabolic network by capturing the stoichiometries of constituent reactions in a stoichiometric matrix, S, and describing a flux configuration as a set of rates at which the reactions in a network are moving (ie the set of reaction fluxes). FBA requires that constraints for some reactions be known, reflecting their maximum or minimum rates. These constraints can either be measured (e.g. uptake reactions) or calculated from physical parameters (e.g. oxygen diffusion) or thermodynamic constraints. In many cases, the constraints can be related to the degree of enzymatic activity for the given reaction. The matrix S and the set of reaction constraints define the set of all possible flux configurations at steady state. A flux configuration can be visualized as a vector in flux space, and all flux configurations that are feasible at steady state lie within a cone in this space, called the flux cone. The core approach of FBA is to choose a metabolic objective which is a linear function of fluxes, and then use linear programming to optimize this objective subject to the constraints. The algorithm results in one or more flux configurations that are optimal for the chosen metabolic goal, and the optimal production capacity of that objective.
FBA provides a method for exploring capabilities and states of a metabolic system at steady state, and genome scale metabolic models can be reconstructed based on annotated genome sequences coupled with literature curation 
. FBA has been used to successfully predict the metabolic phenotype of gene knockouts 
, and the use of metabolic modeling in this case has the advantage of predicting nutrient-dependent phenotypes. FBA has also been used to predict the time courses of growth, substrate uptake, and metabolite production by both Escherichia coli
and Mycobacterium tuberculosis
using a pseudo-steady-state dynamic modeling approach 
. FBA has recently been used as part of an integrated analysis scheme for drug identification; there is a recent publication (targetTB) by Raman et al. that reports this approach 
While powerful, FBA is limited in that it does not take into account the gene regulatory state, as described for example by gene expression data. In effect, the basic approach predicts metabolic capabilities assuming all reactions have the same maximum capacity. Indeed, many of the errors in the prediction of gene knockout phenotype were traced back to the lack of gene regulation in standard FBA models 
. Incorporating a Boolean model of gene regulation with FBA allows the prediction of more biologically realistic dynamic behaviour, including for example a diauxic shift in response to changing carbon source availability 
. However, this approach reduces gene expression to Boolean variables, using either a constant value or 0 for the upper flux bound, rather than making use of direct measurements of gene regulation through whole cell expression data.
We have developed a method, which we call “E-Flux”, to predict metabolic capacity based on expression data. E-Flux extends FBA by incorporating gene expression data into the metabolic flux constraints. We applied E-Flux to M. tuberculosis
), the pathogen that causes tuberculosis (TB). An estimated one third of the world's population has been exposed to this disease, which is estimated to kill 1.6–1.8 million annually worldwide. Multiple drug resistant (MDR) and extensively drug resistant (XDR) strains of tuberculosis are emerging worldwide, so the development of new drugs is of the essence. Bacterial metabolism plays an important role in TB pathology, both in terms of metabolic alterations associated with intracellular growth 
as well as through the production of metabolic products associated with virulence – including mycolic acids 
. Given M. tb
's slow growth rate, the hazards of experimenting directly with this infectious organism, and limitations in measuring all metabolites simultaneously, there is considerable motivation to augment experimental approaches with computational methods for predicting M. tb
We used E-Flux to predict the impact of drugs and environmental conditions on mycolic acid biosynthesis capacity in M. tb, based on a compendium of expression measurements from these conditions. Our method successfully identifies seven of the eight known inhibitors of mycolic acid or fatty acid production that were present in the compendium. E-Flux also correctly predicts whether conditions are directly inhibiting mycolic acid production, or inhibiting production indirectly through other mechanisms. Our method thus provides a promising approach to modeling metabolic state from whole cell measurements of gene regulation.