The problem to be solved is to determine what the flux values are for the different reactions in the human metabolic model.

As far as I understand, a good way to do that would be to use gene expression data to calculate a flux, but I am not sure that is a good idea or if it is even possible. I am not very sure of what is the correct process to do this.

Thanks for any help or suggestions!

  • 1
    $\begingroup$ A metabolic flux is the rate at which a metabolite gets converted into another metabolite by an enzyme. Metabolites aren't coded for by genes, so gene expression data doesn't help determine metabolic flux, unless there is some feedback that will cause the concentrations of the various enzymes involved to change via gene expression. But that occurs on a timescale which is much slower than typical metabolic flux. $\endgroup$ – A. Kennard Feb 17 '14 at 2:20

I am no expert on the field, but quantifying gene expression isn't a simple problem, although it is a possibility.

A solution I know to be used is using radioactive markers. See for example this review and particularly "The cell-cell lactate shuttle" section, which might give some insights.

This can include radioactive tracers (such as [14C]lactate), non-radioactive tracers (with [13C] or [2H]), "net exchange measurements" (I don't really know what this means here, perhaps the kind of exchange calculations they present here), and muscle biopsies (it is then possible to measure the metabolite concentration by enzymatic reaction or spectrophotometry for example).


Summary: Try one of the extensions of FBA which take gene regulation into account, but be aware of the limitations. See below for references.

Long answer:

There are two approaches to estimating internal metabolic fluxes: The (primarily) experimental and the (primarily) computational. I say primarily, because even for the experimental approach, a large amount of computation need to be done, and for the computational approach, experimental studies must be consulted in order to set realistic model parameters.

Experimental determination of internal metabolic fluxes has been centered around 13C-metabolic flux analysis (13C-MFA) performed through carbon-labelling experiments, as mentioned above. 13C-MFA is complicated and expensive, so there is a relative lack of experimental data, and most experiments cover only a few fluxes when compared to the number of possible reactions, typically several thousand in genome-scale metabolic models. In 13C-MFA, most experiments estimate the fluxes in a much smaller model (for example the central carbon metabolism), avoiding or ignoring other possible reactions, and the set of fluxes that are estimated varies between experiments.

Models which are used in computational approaches to flux analysis are typically larger than the models used in 13C-MFA experiments, and thus it is usually not possible to determine all the fluxes from experimental data alone. Thus, an approach called constraint-based modelling (COBRA) is mainly used. The basic assumption of most current flux analysis methods, both experimental and computational, is that metabolite concentrations are at steady state. Mathematically, this can be described by the equation

$S\overrightarrow v = 0$

where $S$ is a stoichiometric matrix relating the metabolites and all possible reactions (basically a compact description of the metabolic model), and $\overrightarrow v$ is the flux vector containing all the flux values. In addition to the steady state requirement, constraints are typically applied to the uptake and excretion rates of various metabolites, limiting them to biologically realistic values. Other requirements, such as requirements for ATP maintenance consumption can also be applied.

Because there are typically many possible flux patterns that obey the above constraints, a principle is needed for selecting a biologically realistic solution from the solution space of feasible flux patterns. Assuming that cells optimize their metabolic patterns in some way, different objective functions are used that attempt to capture the metabolic behaviour for cells. Once an objective function is chosen, the set of possible solutions can be searched for the flux pattern that gives the highest objective value as a function of the flux vector. Given that the objective is linear (simply a weighed sum of fluxes), an optimal solution can be found rapidly. In general, many different optimal solutions may exist for a given objective function. The method of applying an objective function to a constrained metabolic model at steady state is called Flux Balance Analysis (FBA).

The most basic and a popular objective is maximization of biomass production. While useful for determining maximal growth rates, this objective is unlikely to be realistic when applied to human cells. FBA and related methods is generally well-suited to determining performance limits for single endpoints such as growth rate or production of a single metabolite, but not capable of precisely determining all internal metabolic fluxes. The point that many optimal solutions may exist for a single objective function is important in this regard.

As 13C-MFA and FBA are based on the same concept of metabolite balancing, they can be viewed as different ends of the same spectrum, from purely experimental to purely computational, with 13C-MFA constrained FBA (where 13C-MFA results is used to constrain the possible fluxes in a model before optimizing an objective function, or minimizing the difference between the experimental fluxes and the FBA solution, subject to additional constraints) in the middle.

For an introduction to Flux Balance Analysis, see Orth, Thiele & Palsson: "What is flux balance analysis?" Nature Biotechnology 28 245-48 2010.

For performing Flux Balance Analysis, several software packages are available. One of the most used is the COBRA Toolbox for Matlab. A version for Python called CobraPy is also under development. Both are available at http://opencobra.sourceforge.net/openCOBRA/Welcome.html

Note that most gene-expression data only shows relative changes in expression between two conditions. Most methods are thus based on comparing two different conditions. More recently, RNA-sequencing (RNAseq) may also be used to obtain more direct measures of gene expression levels. Many extensions and variations on FBA have been published which take into account gene expression. Some of these are regulatory FBA (rFBA, Covert & Palsson: Journal of Biological Chemistry, 2002 277, 28058-28064), Metabolic Adjustment by Differential Expression (MADE, Jensen & Papin: Bioinformatics. 2011 Feb 15;27(4):541-7 ) and iMAT (Schlomi et al: Bioinformatics (2010) 26 (24): 3140-3142.). For a more recent method, gx-FBA (gene expression-FBA), see *Navid & Almaas, BMC Systems Biology 2012, 6:150).

You may also want to read this article which deals with construction tissue-specific metabolic models: Computational reconstruction of tissue-specific metabolic models: application to human liver metabolism. Mol Syst Biol. 2010 Sep 7;6:401. doi: 10.1038/msb.2010.56.

The problem with using FBA and related methods to estimate internal fluxes is that validation of the results is difficult because of the mentioned lack of experimental data. I wrote a project report on the topic during the last year of my Master's, which includes a basic description of 13C-MFA. It can be read at http://www.slideshare.net/jarlemag/rapport-31295058

  • $\begingroup$ This is a wonderfully complete answer. Thank you very much! I am actually on the path of using FBA to achieve this. In our project, we have several cancer cell lines from the CCLE that we are trying to simulate through FBA. However, we would like to "customize" the Human Metabolic Model to the characteristics of each cell line in order to get a better baseline for analysis. The CCLE provides some gene expression data, so we'll definitely look into that. Again, thank you! If you have any other suggestions, I would gladly hear them :). +1. $\endgroup$ – Juan Carlos Coto Feb 17 '14 at 21:10
  • 1
    $\begingroup$ Glad to be of help. It's a bit difficult to give suggestions without knowing more specifically what you want to do, but feel free to ask if you have any more questions. May I ask what software you will be using for the analysis? $\endgroup$ – jarlemag Feb 17 '14 at 21:19
  • $\begingroup$ Absolutely. So far, I have chosen COBRAPy to run FBA (don't have access to MATLAB and I'm experienced with Python). Perhaps the greatest aid I could get from you would be your opinion. I have asked this about using the HMM, as well as this on where to get good HMM data. In essence, I am trying to gauge what I need to do to predict the outcome of gene knockouts by cancer therapy drugs. $\endgroup$ – Juan Carlos Coto Feb 17 '14 at 21:31
  • 1
    $\begingroup$ That's cool. I'm in the same situation, also using CobraPy. I like it better than the MATLAB Toolbox, but unfortunately it doesn't have as many modules available. I've been working on implementing some methods only published as MATLAB code (or not at all) for use with CobraPy, but they're not quite ready yet. I'll see what I can do for your other questions, although I haven't worked on any applications of FBA to human metabolism. $\endgroup$ – jarlemag Feb 17 '14 at 21:43

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.