Biology Reference
In-Depth Information
falsely predict a greater growth rate than would be expected
experimentally.
Phenotypic Tests
Once topological concerns and loops have been addressed,
qualitative tests are conducted to assure that the model can
make important metabolites and to identify which reactions
cannot carry flux.
One important test of a network reconstruction is its
ability to produce all essential biomass components and
experimentally measured secretion products. When FBA
predicts zero growth or no product secretion, non-producible
metabolites are identified by using linear programming to
maximize flux through demand reactions for each individual
secretion product or reactant in the biomass function.
Metabolites that cannot be synthesized according to the
model are remedied through additional gap-filling, in which
additional reactions are suggested that would allow synthesis
of themetabolite to occur
[9]
. Since this process is sensitive to
the choice of model growth medium, it is repeated multiple
times for different in silico growth conditions.
Just as some biomass and secretion metabolites might
not be produced, it is common for some reactions to be
unable to carry flux. These reactions involve metabolites
that are solely produced or consumed or that depend on
other non-functional reactions. These can arise due either to
knowledge gaps (when the connecting reactions are
unknown) or to scope gaps if subsequent metabolic steps
extend into a system outside the scope of the model. For
knowledge gaps, gap-filling is done by adding the missing
reaction, or a proposed reaction if none is known. However,
scope gaps are better addressed with the use of sink and
demand reactions, if doing so improves model accuracy for
its intended purpose.
Once gap-filling is done, the comparison of phenotypic
screens
[49]
with in silico gene deletion studies allow for
a detailed assessment of the accuracy of functional reac-
tions within the network. In this, single-gene deletion
phenotypes (e.g., growth rate) are computed to validate the
model in a holistic sense. When model-predicted growth
differs from experimental growth phenotypes, these fail-
ures guide further curation. If the model predicts growth of
a gene deletion mutant, whereas experimental data show
the gene deletion to be lethal, then it is possible that
a reaction has been improperly incorporated, that a gene
associated with another reaction is silenced, or that the
biomass reaction is incomplete. If, however, the model does
not grow despite measured experimental growth, it is likely
that reactions are missing from the reconstruction, and
therefore gap-filling is needed
[38]
. It is important to note
that the lack of regulatory rules (e.g., transcription factor
activation/repression) in the model can also result in
improper predictions. For example, growth in a medium
with two carbon sources in it will not proceed in a diauxic
[50]
fashion as expected experimentally. Rather, the model
will consume both carbon sources simultaneously and
Quantitative Tests
The ultimate goal of a genome-scale reconstruction is to be
able to use it to make clear quantitative predictions that
yield novel insight and knowledge. Therefore, it is critical
to test the model to ensure its quantitative accuracy in
predicting physiological parameters. Such quantitative tests
may include predicting growth rate
[51]
or P/O ratio, which
is a measure of ATP generated per electron pair used in
oxidative phosphorylation
[52]
.
For growth rate predictions, the biomass function is
optimized under a variety of media conditions and subse-
quently compared with experimentally measured growth
rates. Incorrect in silico predictions may include (1) no
growth when growth is expected, (2) slower growth than
physiologically expected, (3) faster growth than physio-
logically expected, or (4) growth when no growth is
expected. These different scenarios have several possible
explanations, and the appropriate steps to take for each
scenario are discussed below.
In the first scenario, when the model predicts no growth
while experimental assays demonstrate growth, it means
that the network is incomplete (
Box 12.7
). Therefore, the
reconstruction requires additional gap-filling and the
accuracy of constraints (e.g., metabolite uptake or biomass
function composition) needs to be verified.
BOX 12.7 Gap-Filling in the Core
Model
Consider the in silico gene deletion of b3919, which codes
for the triose phosphate isomerase protein that participates in
glycolysis. Simulation of the knockout using the core meta-
bolic model predicts that the deletion of this gene is lethal.
However, experimentally, this mutant is known to exhibit
growth. Investigation into the cause behind the false
prediction shows that dihydroxyacetone phosphate (DHAP)
can be produced but not consumed. Thus, the model predicts
that the mutation is lethal because DHAP would accumulate.
This fact suggests that there is possibly a missing reaction that
would consume DHAP and prevent its accumulation in the
knockout. In this case, the reactions necessary to account for
DHAP utilization (e.g., methylglyoxal synthase) were not
included in the reconstruction because they fell outside the
scope of the reconstruction target of core metabolism.
Various methods of correction exist, but as the pathway
needed is outside the scope of the reconstruction, addition of
a demand reaction for DHAP may be an appropriate course
of action. Simulation of this modified version of core
metabolism is able to make the proper qualitative prediction
of growth for the triose phosphate isomerase mutant.
E. coli
