Genome-Scale Models of Metabolic and Regulatory Networks - Systems Biology: Networks, Models, and Applications - page 244

studies described above are the incorrect predictions. Detailed analysis
of incorrect predictions allows identifying potential changes to the
in silico models and in general increasing our understanding of the
metabolic physiology of the organism [19].
In addition to the standard FBA approach, the MoMA and ROOM
methods described above have also been used to perform large-scale
gene deletion studies [38,39]. In general, these methods perform simi-
larly to FBA in predicting qualitative growth phenotypes. However,
both MoMA and ROOM predict experimentally measured intracellular
flux distributions for nonevolved knockout strains significantly better
than FBA. This result indicates that in vivo organisms are indeed lim-
ited in their ability to adapt to genetic changes. The FBA approach,
which assumes full optimality for the knockout strains, tends to over-
estimate the degree of adaptation allowed.
ROBUSTNESS AND REDUNDANCY
The applications described above illustrate only the ability of genome-
scale metabolic models to predict directly experimentally measurable
quantities such as growth rates. In addition to these direct predictions,
metabolic reconstructions can also be studied using, for example,
flux coupling or extreme pathway analysis. The results from these
analyses are primarily intended to address global network-level ques-
tions such as the relationship between metabolic network structure
and robustness to genetic and environmental changes [60,61]. However,
the insights obtained in this way on the global organization of meta-
bolic networks can be used to interpret indirectly experimental data
sets such as synthetic lethality data [62,63] and to derive experimen-
tally testable hypotheses.
Extreme pathway analysis has been used to study the built-in
redundancy in metabolic networks in the form of the number of
distinct routes a metabolic network can use to produce a given set of
outputs from a given set of inputs. For example, comparison of extreme
pathway redundancy for amino acid biosynthesis between H. pylori
and H. influenzae indicated a much smaller degree of redundancy in
H. pylori [61,64]. This decreased redundancy has implications for the
robustness of the amino acid biosynthesis function with respect to
genetic changes. Pathway redundancy measured through elementary
flux modes has also been shown to correlate with quantitative gene
deletion phenotypes in E. coli central metabolism. This analysis
indicated that the pathway-based redundancy metrics can indeed
predict in vivo robustness and redundancy [60].
An alternative way to study redundancy in metabolic networks is
to enumerate the minimal reaction sets required for in silico growth
under various growth conditions [65]. This enumeration task can be
reformulated as a mixed-integer optimization problem where the