Biology Reference
In-Depth Information
In the second scenario, in silico growth rates are slower
than experimental measurements. This can occur if one or
more components are not being synthesized at a high
enough rate. Thus, the limiting components are identified
by iteratively supplementing the model with each biomass
component and measuring its effect on the predicted
growth rate. This approach provides insight into which
pathways require more careful curation.
The third scenario, in which model-predicted growth
rates are higher than experimentally measured rates, can be
explained by several means. One possibility is that the cell
may have additional objectives beyond the synthesis of
biomass
[53
e
60]
. Thus, when the model assumes the
optimization of growth, it overestimates the amount of
material and energy equivalents used for biomass produc-
tion. Another possibility is that GAM or NGAM may have
been improperly calculated for that growth condition.
Overestimation of growth might also result from improp-
erly included reactions. It is also recommended to constrain
directionality of reactions that consume ATP or use
quinones as electron acceptors to prevent these reactions
from running in reverse and improperly increasing biomass
reaction components. Model reactions that facilitate the
erroneous high growth rate may be identified by testing the
sensitivity the predicted growth rate to changes in flux for
each reaction. Lastly, it is also possible that regulatory
mechanisms suppress gene expression or enzyme activity
in vivo. Thus, the experimental data may represent a non-
ideal phenotype. Previous adaptive laboratory evolution
(ALE) experiments have demonstrated that after hundreds
of generations of exponential growth, bacterial strains can
adapt towards the model-predicted growth rate
[34,61
e
63]
.
In the fourth scenario, the model predicts that the
organism can grow when experiments say otherwise. This
may result from the improper inclusion of reactions, lack of
accounting of regulatory mechanisms suppressing growth
in vivo, or the presence of sink reactions that allow growth
without consuming the metabolites in the medium. Careful
curation, comparison with expression data, and iterative
testing following sink adjustments can help reconcile such
model
e
phenotype inconsistencies.
improvements on the scope of the reconstruction, as has
been done for the genome-scale model for E. coli
(improving coverage from 660
[70]
to 1366
[71]
genes) and
Saccharomyces cerevisiae
[72
e
74]
. Moreover, the model
can be used for a plethora of applications
[75]
, such as the
examples discussed later in this chapter.
RECONSTRUCTION STANDARDS
So far we have discussed the general concepts to be
considered in a network reconstruction. However, it is
important to follow all of the detailed steps in the pub-
lished reconstruction protocol
[9]
to ensure a high-quality
reconstruction and accurate model predictions. Failing to
carefully curate a reconstruction can lead to a host of
errors, such as the incorporation of incorrect reactions into
the reconstruction, the loss of reactions that were missed
in the draft reconstruction due to incomplete genome
annotation, and inaccurate simulations due to incorrect
stoichiometry (e.g., if protons or water are not added to
balance reactions, processes such as oxidative phosphor-
ylation will be affected). Thus, without curation through
organism-specific literature or experimental data, mistakes
of this sort will be pervasive. In addition, failing to vali-
date the reconstruction computationally will result in
a model with low predictive power that may not be able to
simulate known phenomena.
APPLICATIONS
Since their development, genome-scale metabolic models
have been deployed for a plethora of different applications
and analyses. Many applications easily fit within one of six
categories (
Figure 12.5
)
[75,76]
. Each category is described
here and relevant examples are discussed.
Metabolic Engineering
The genetic engineering of organisms to produce
commodity chemicals and pharmaceutical compounds
more efficiently has been of interest in the life sciences for
decades
[77]
. Prior to the advent of genome-scale models,
metabolic engineering was approached on a small scale
with targets based on intuitive changes in metabolic path-
ways closely connected with the product of interest
[76]
,an
approach that has been successful for certain applications
[78]
. However, this approach does not allow for the iden-
tification of many non-intuitive changes. Approaching
metabolic engineering from a systems perspective (i.e.,
using genome-scale models) allows system-wide conse-
quences of genetic manipulations to be predicted. In
addition, these approaches can expand the list of potentially
beneficial manipulations one can attempt.
Stage 5: Data Assembly, Dissemination,
and Use
Once a network reconstruction is completed, it is frequently
made available to the research community. Often all of the
relevant information, curator comments, and citations are
presented in a spreadsheet, and the model is released in
SBML format
[64]
for use with an array of modeling
programs
[65
e
69]
. Important modeling parameters should
be presented in an organized manner, thereby allowing for
the replication of the validation tests. Careful efforts of
model preparation and dissemination will allow iterative
