Available via license: CC BY 4.0
Content may be subject to copyright.
Contents lists available at ScienceDirect
Computational and Structural Biotechnology Journal
journal homepage: www.elsevier.com/locate/csbj
Mini-Review
Genome-scale models as a vehicle for knowledge transfer from microbial
to mammalian cell systems
Benjamin Strain, James Morrissey, Athanasios Antonakoudis, Cleo Kontoravdi
⁎
Department of Chemical Engineering, Imperial College London, London SW7 2AZ, United Kingdom
article info
Article history:
Received 15 November 2022
Received in revised form 6 February 2023
Accepted 6 February 2023
Available online 8 February 2023
Keywords:
Mammalian cell metabolism
Resource allocation models
Flux balance analysis
Human pathophysiology
abstract
With the plethora of omics data becoming available for mammalian cell and, increasingly, human cell
systems, Genome-scale metabolic models (GEMs) have emerged as a useful tool for their organisation and
analysis. The systems biology community has developed an array of tools for the solution, interrogation and
customisation of GEMs as well as algorithms that enable the design of cells with desired phenotypes based
on the multi-omics information contained in these models. However, these tools have largely found ap-
plication in microbial cells systems, which benefit from smaller model size and ease of experimentation.
Herein, we discuss the major outstanding challenges in the use of GEMs as a vehicle for accurately analysing
data for mammalian cell systems and transferring methodologies that would enable their use to design
strains and processes. We provide insights on the opportunities and limitations of applying GEMs to human
cell systems for advancing our understanding of health and disease. We further propose their integration
with data-driven tools and their enrichment with cellular functions beyond metabolism, which would, in
theory, more accurately describe how resources are allocated intracellularly.
© 2023 The Authors. Published by Elsevier B.V. on behalf of Research Network of Computational and
Structural Biotechnology. This is an open access article under the CC BY license (http://creative-
commons.org/licenses/by/4.0/).
1. Introduction
Genome-scale metabolic models (GEMs) are a comprehensive
representation of the link between genotype and phenotype, sum-
marising information on genome, proteome and metabolome of a
cell [1]. This information is organised in the form of matrices relating
genes to metabolic reactions and reactions to metabolites, as well as
a set of gene-protein-reaction (GPR) associations [2,3], as shown in
Fig. 1. The construction of the matrices and GPR associations relies
on genomic, transcriptomic, proteomic and metabolomic data [4].
Given this information and a set of metabolite uptake/secretion rates
that act as constraints, GEMs can calculate the rates of intracellular
reactions, thus providing fluxomic information.
GEMs have been constructed for over 6000 organisms [5,6], in-
cluding the well-studied Escherichia coli [7], Mus musculus [8,9], Pi-
chia pastoris [10], Saccharomyces cerevisiae [11] and Homo sapiens
[12–14], with reconstructions regularly updated to include more
complete GPR associations and remove blocked reactions and dead-
end metabolites. GEMs can be used to study cell metabolism, opti-
mise bioprocesses, and design strains with enhanced or custom
functionality. Historically, GEMs have found greater application in
microbial organisms, owing to the smaller model size and relative
ease of experimental validation/manipulation compared to mam-
malian cell systems. The smaller metabolic network size of microbial
model systems such as, for example, E.coli cells, has also led to the
development of a variety of solution methodologies and optimisa-
tion algorithms for the design of cells with desired phenotype
(summarised in [15]). Although the transfer of the entire repertoire
of techniques to mammalian cell systems is often hampered by in-
creased model size and complexity leading to highly under-
determined models, there are already developments in the use of
mammalian cell GEMs for strain and process engineering (e.g.,
[16–18]), as well as recent algorithm development work applied for
understanding the nutritional needs of bioprocessing-relevant or-
ganisms but also Atlantic salmon (Salmo salar) [19].
In this work, we outline the main remaining challenges in the
application of GEMs and related toolkits to mammalian cell systems
and zoom in on their application to human cells as a vehicle for
understanding health and disease.
2. On the use of GEMs for understanding human cell systems
Advances in clinical sample analysis and in vitro disease models
are now enabling the generation of similar datasets for human
Computational and Structural Biotechnology Journal 21 (2023) 1543–1549
https://doi.org/10.1016/j.csbj.2023.02.011
2001-0370/© 2023 The Authors. Published by Elsevier B.V. on behalf of Research Network of Computational and Structural Biotechnology. This is an open access article under the
CC BY license (http://creativecommons.org/licenses/by/4.0/).
]]]]
]]]]]]
⁎
Corresponding author.
E-mail address: cleo.kontoravdi@imperial.ac.uk (C. Kontoravdi).
physiology and pathophysiology. It is therefore opportune to ex-
amine how learnings and techniques developed for analysing data
from biotechnologically relevant organisms using GEMs can be ap-
plied in a clinical context to further our understanding of health and
disease (Fig. 2). For example, within industry, it is commonplace to
generate GEMs specific to a cell line via the integration of ‘omics data
to prune inactive reactions using techniques such as GIMME [20]
(cell line specific model generation is reviewed in depth in [21]). The
same holds true in health and disease research, with thousands of
patient-derived GEMs having been published for cancer alone [22].
As these models are specific to each disease type, they can be ef-
fectively used to explore essential genes in diseased tissue and to
identify drug targets. In a recent study, for instance, single-gene
knockouts were performed on GEMs of NCI-60 cancer cell line panel
to identify and rank genes responsible for the growth of cancerous
cells in an effort to identify potential drug targets that would reduce
the growth rate of cancer cells but not that of normal cells [23]. This
type of analysis is not only limited to chronic diseases; it has also
been used in infectious disease studies. In one body of work, flux
balance analysis (FBA) was applied to human lung cells infected with
severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) and
host-specific essential genes and gene pairs were determined
through in silico knockouts that were theorised to reduce viral bio-
mass production without affecting the host biomass [24].
These examples highlight the potential of GEMs for analysing
large clinical omics datasets in a systematic way that links multiple
levels of information. In our opinion, it is possible to envisage the
use of GEMs and related methodologies, such as strain design al-
gorithms, in a health setting to generate optimal strategies that re-
duce disease-associated phenotypes, while improving desirable
healthy phenotypes. Herein, we outline the main outstanding chal-
lenges towards this end.
2.1. Challenges when building mammalian GEMs
Eukaryotes are known to be more biologically complicated that
prokaryotes, meaning that applying GEM techniques to eukaryotic
organisms is more challenging with respect to obtaining accurate
predictions. One of the key sources of difference in complexity be-
tween prokaryotes and eukaryotes is the presence of subcellular
organelle structures, such as mitochondria, peroxisomes, and nu-
cleus, that do not exist in prokaryotes. Any well annotated eu-
karyotic GEM must contain these structures and the reactions
associated with them for truly accurate predictions [25]. Sig-
nificantly however, the presence of sub compartments means there
is a requirement to gap fill the model using intracellular transport
reactions. These transport reactions are often poorly studied and can
lead to models with many reversible reactions, which, in turn, may
lead to futile cycles, freely exchanging metabolites and protons
across compartments, and erroneous energy generation calculations
[26]. These cycles have been shown to inflate maximal biomass
production rates by 25 % and are known to be present in the majority
of published genome scale models [27], with eukaryotic models at
greater risk thanks to the increased presence of intracellular ex-
changes. Ultimately, these results highlight the importance of using
an appropriate combination of gap filling algorithms (reviewed in
depth here [28,29]) and manual curation when moving from pro-
karyotic to more complex, eukaryotic models of metabolism to en-
sure accurate predictive performance.
While GEMs have been developed for many different species across
all domains of life (reviewed in [6]), given the complexity of building
eukaryotic GEMs, there has been a lack of regularly updated and
publicly curated GEMs for mammalian model organisms such as Mus
musculus (mouse) and Rattus norvegicus (rat) [30]. Instead, recent
research developing new modelling techniques using non-human
Fig. 1. Reconstruction of a generic GEM from the different layers of ‘omics datasets’.
B. Strain, J. Morrissey, A. Antonakoudis et al. Computational and Structural Biotechnology Journal 21 (2023) 1543–1549
1544
mammalian GEMs has predominantly focused on industrially relevant
organisms such as Chinese Hamster Ovary (CHO) cells. To help address
this, a framework has recently been published that combines multiple
data sources, including the Kyoto Encyclopaedia of Genes and Genomes
(KEGG) [31], and generates a coherent collection of GEMs for major
model animals using the Human1 GEM as a template [30,32]. This
approach allows for the straightforward development and main-
tenance of GEMs for multiple species. Since small rodents account for
90% animals used annually in medical research [33], the development
of these models using a high-quality model as a backbone opens the
possibility to better utilise GEMs in medical research settings, reduce
the reliance on model animals and understand differences between
model animal and human metabolism.
2.2. Determining effective exchange reaction constraints
One of the first challenges that occurs when running GEMs is
gathering sufficient extracellular metabolomic data to effectively
calculate metabolite uptake rates to constrain the GEM of interest.
Within industrial biotechnology, the calculation of these uptake
rates is straight-forward, thanks to the relative ease at which ex-
tracellular metabolomics can be measured in bioreactors. This
means that industrially relevant cultured mammalian cells, such as
CHO cell lines, often have detailed constraints for many exchange
reactions. This has allowed researchers to understand how the ac-
curacy of this data can affect GEM predictions. For instance, using
the CHO cell GEM, researchers have demonstrated that the mea-
surement of low exchange rates of essential amino acids has the
biggest impact on the growth rate prediction [34] and that the highly
accurate quantification of all uptake and secretion rates was essen-
tial for reliable predictions generated by FBA [35].
The generation of such extracellular time-course metabolomics is
far more challenging in multicellular organisms. While researchers
can culture the cells of interest in vitro, this may not be fully re-
presentative of how a tissue behaves in vivo. This means that gen-
erating in vivo constraints is of vital importance to accurately
understand diseased states, toxicology, and nutrition. A potential
method to do this is the use of nutrition databases to calculate the
approximate composition of metabolites in a diet that are available
for uptake by a cell. One such database is the Virtual Metabolic
Human [36], which contains the composition of 11 pre-defined diets
that can be downloaded as a flux rate (in mM per person per day).
This data can be directly used to constrain the human metabolic
model. Significantly, while this resource acts as an excellent baseline
for constraining the human GEM to understand differences in diet,
given that small changes in the exchange rates of essential amino
acids can significantly impact the accuracy of predictions in the CHO
GEM [35], it seems unlikely that such a database would provide
enough accuracy to consistently give meaningful outputs from a
GEM in all use cases.
To overcome this obstacle, techniques that rely on true in vivo
measurements, such as arterio-venous blood metabolomics (AVBM)
profiles, may be considered. In this approach, blood samples are
taken from an artery directly before and a vein directly after the
tissue type of interest. The difference in metabolite concentrations
between these two samples is then presumed to be the amount of
metabolite exchanged by the tissue of interest, which can be used to
constrain the GEM. This approach has recently been applied to the
genome scale modelling of multi-cellular organisms. In one body of
work, researchers used AVBM measurements to constrain a GEM to
study the global metabolism of liver and intestine of a minipig
model of obesity, leading to the identification of upregulated path-
ways in obese subjects, such as tryptophan metabolism [37].
Nonetheless, while this approach may be appropriate for the
genome scale modelling of animal models in the lab, it is highly
invasive and unlikely to be acceptable for humans.
Fig. 2. Summary of areas where knowledge can be transferred from mammalian cell GEMs to wider health and disease research.
B. Strain, J. Morrissey, A. Antonakoudis et al. Computational and Structural Biotechnology Journal 21 (2023) 1543–1549
1545
2.3. Determining appropriate objective functions
The common selection of a biomass maximisation as an objective
function for performing FBA of mammalian GEMs is a methodology
that largely remains from microbial GEMs, despite that fact it is well
known to not be representative of the true ‘objective’ of a mam-
malian cell, especially outside the exponential growth phase [38].
This lack of suitability of a biomass objective is even more apparent
for in vivo systems where, unless the tissue of interest is cancerous,
cells rarely maximise their proliferation. As a result, researchers
trying to model in vivo systems must consider the use of alternate
objective functions and draw inspiration from mammalian bio-
technology solutions. For instance, an unconventional objective
function based on the minimisation of non-essential nutrient uptake
has been designed for the CHO cell GEM [39]. This method directly
estimates essential amino acid uptake fluxes by solving for the
“essential minimum” consumption requirements based on cellular
growth measurements. This unconventional objective function was
shown to distinguish metabolic differences between three distinct
CHO cell lines not directly observed using the conventional biomass
maximisation. This highlights how the use of more appropriate ob-
jective functions may render GEM outputs more information rich,
improving their practical application in health and disease research.
The identification of more appropriate objectives may either be
achieved applying well established knowledge around the tissues of
interest (e.g., a GEM of a B cell may be set to maximise antibody
production) or by inferring cell functions through data, such as the
analysis carried out by Richelle et al. [40]. In this work, the functions
of a cell were inferred from transcriptomics data by considering the
gene expression level associated with a metabolic pathway and the
number of reactions involved. During this work, a list of tasks was
curated resulting in a collection of 210 tasks covering seven major
metabolic activities of a cell (energy generation, nucleotide, carbo-
hydrates, amino acid, lipid, vitamin & cofactor and glycan metabo-
lism). These tasks were used to protect selected metabolic features
using context-specific model generation algorithms for human, CHO,
and mouse cell GEMs. The results highlight that these context-spe-
cific models better capture the actual biological variability across cell
lines. Similar methodologies can therefore be considered when
trying to determine the ‘goal’ of a tissue when selecting an objective
function.
In addition to the lack of suitability of maximising biomass, it is
important to consider that the biomass formation of mammalian
cells is highly variable, depending on factors such as environmental
conditions, cell type or culture phase, meaning the biomass equation
must be customised for optimal model performance. For example,
research in CHO cells has demonstrated cell lines display highly
variable total protein content, cell dry mass and lipid composition
across cell lines [35]. Moreover, work using the human GEM showed
that metabolite composition and associated coefficients of the bio-
mass function had a large impact on the growth rate prediction
accuracy of cancer cell lines. In addition, metabolite composition of
the biomass equations significantly impacted gene essentiality ac-
curacy [41], meaning a new biomass equation should arguably be
determined in each case. To this end, tools originally designed for
microbial systems may be used, such as BOFdat [42], to generate
custom biomass reactions for mammalian cell systems based on
experimental ‘omics data.
2.4. On the integration of data-driven modelling with GEMs
In recent years, advances in artificial intelligence and machine
learning have revolutionised many areas of biological research [43].
Such approaches have started to be coupled with GEMs to help
improve predictions and aid model output analysis. The coupling of
GEMs with data- driven methods has been proposed as a method to
effectively reduce the solution space by predicting biologically re-
levant constraints from experimental data (reviewed in depth in
[15,44,45]). As with the previously discussed methodological areas
of genome-scale modelling, this coupling of machine learning with
GEMs to improve predictions is at a more advanced stage in bio-
technologically relevant mammalian cell systems than it is in human
health and disease research. For instance, a recently published
method, termed HybridFBA, coupled unsupervised machine learning
with a CHO cell GEM. In this approach additional flux constraints
were deduced by Principal Component Analysis (PCA) of experi-
mental flux data [46]. Specifically, the authors used each principal
component to impose a constraint on the direction of variation of
groups of fluxes. This method was shown to significantly improve
growth rate predictions compared to standard FBA and was used to
design a culture feed in silico that led to desired phenotype from
target cell lines. This highlights how the coupling of mammalian cell
GEMs with machine learning algorithms can improve their perfor-
mance.
In addition, machine learning methods may be used to better
analyse outputs and extract meaning from complex model predic-
tions. For example, flux distribution predictions may be analysed
using supervised and unsupervised machine learning methodologies
to pick apart key aspects of metabolism that may influence a dis-
eased phenotype of interest. This methodology has already been
well applied within health and disease research using GEMs
[15,44,45]. For example, researchers have used unsupervised
learning with GEMs to identify the fluxes that explain most of the
data variation in breast cancer patients, reduce dimensionality and
create patient groupings [47]. Furthermore, researchers have applied
personalised FBA models of patient tumours to predict metabolite
production rates. These were input into machine learning classifiers
for the identification of metabolite biomarkers associated with ra-
diation resistance. The results demonstrated improved classification
accuracy and identification of clinical patient subgroups, marking a
significant step toward personalised classifiers for radiation treat-
ment response [48]. These approaches demonstrate the power of
using these two techniques synergistically.
3. A case for resource allocation models
3.1. Benefits of resource allocation
There has been a drive in the systems biology community away
from classical stoichiometric network study and towards the study
of metabolism through an optimised cellular economy. Resource
allocation models (RAMs), as recently reviewed in [49–52] and with
key methods summarised in Table 1, can describe many aspects of
metabolism and cellular behaviour [53,54], where simple stoichio-
metric balances fall short. So far this drive towards RAMs has been
almost exclusively carried out in microbial systems, due to their
relative simplicity. Enzyme constrained FBA (ecFBA) models are also
considered in this review due to their similarities with RAMs and are
included in Table 1.
One key advantage of a RAM approach to metabolic modelling is
that the additional constraints greatly reduce the feasible solution
space by placing more restrictive bounds on fluxes. This lowers the
variability of metabolic fluxes and guides flux towards more biolo-
gically feasible solutions. This would be particularly useful for
mammalian GEMs, which contain many thousands of reactions
[9,55,56][refs], and hence have extremely large potential solution
spaces.
As well as reducing the solution space of metabolic models, the
additional constraints also predict and explain key phenotypes that
are not possible with traditional stoichiometric models [57–60].
Classical models ignore costs related to synthesis and usage of
proteins and are limited only by the stoichiometry of metabolites
B. Strain, J. Morrissey, A. Antonakoudis et al. Computational and Structural Biotechnology Journal 21 (2023) 1543–1549
1546
exchanged by the cell with its surroundings, meaning, even if a re-
action is unlikely to occur due to the production of an expensive
catalysing enzyme, the model is unable to account for this. The
communal usage of resources drastically effects the distribution of
fluxes through the model. Phenomena such as overflow metabolism
do not make sense from a purely stoichiometric point of view and
can only be explained in the context of the trade-off between ‘in-
efficient’ metabolism, protein cost and cell growth [58,61,62]. The
ability to predict overflow metabolism is an important feature of
mammalian cell modelling, such as the Warburg Effect in cancer
cells. Being able to better predict peripheral overflow metabolism
would be beneficial in the metabolic modelling for clinical research
of diseases such as cancer, where non central pathways are known to
play a key role [63–66].
In addition to cancer cell biology, overflow metabolism is im-
portant in biopharmaceutical production using mammalian cells e.g.,
CHO cells. CHO cells typically undergo a lactate-producing phase, in
which overflow metabolism is high, followed by a lactate consuming
phase as growth rate subsides [67]. The accumulation of lactate is
toxic to cell cultures, causing the addition of base to maintain pH set
point and subsequently raising osmolality and lowering growth rates
[68,69]. The ability to accurately capture lactate producing and
consuming phases through metabolic modelling would aid in pro-
cess and cell line optimisation.
More recently, other phenomena have been effectively modelling
through proteome allocation, for example arginine catabolism in L.
lactis [70]. The application of resource allocation to mammalian
metabolism would be able elucidate features that have yet to be
observed in traditional metabolic modelling.
A further benefit of expanding classical models with resource
allocation machinery is the ability to incorporate omics data more
effectively. With the increased availability of omics data, GEMs
provide an excellent framework for the integration of this data into a
combined workflow. As RAMs can consider transcription and
translation machinery, transcriptomics and proteomics can be used
to constrain metabolism in a more targeted manner, as opposed to
current methods which rely on assumptions on the link between
reaction rate and gene expression/protein translation [71–73].
The broadened scope of RAMs allows a more complete under-
standing of cell behaviour and the relationship between cellular pro-
cesses. This allows predictions that could not be captured with classical
models, such as identifying bottlenecks and gene engineering targets as
well as biological parameters e.g., condition-dependent biomass com-
position [74,75] and transcription/translation machinery [74].
3.2. Challenges in implementation to mammalian systems
While the benefits of RAMs and ecFBA in mammalian systems are
numerous, there are obstacles on the path to achieving this goal. One
of the main challenges is the scarcity of enzyme data. EcFBA, in
particular, rely on the choice of turnover number (k
cat
) values, which
are difficult to source for mammalian cells. For example, Yeo et al.
were able to find k
cat
values for 16 % of enzymes in their CHO GEM
[76], and several of these were taken from other organisms (e.g.,
rodent and human) when there was no Chinese hamster data
available. Additionally, in vitro k
cat
measurements may differ from
those in vivo, although the two have been shown to be correlated
[77]. These factors render the application of ecFBA to mammalian
cell systems difficult and prevent their full utilisation. A potential
solution is to use machine learning approaches for k
cat
prediction
[78], which the enzyme amino acid sequences and the structures of
their substrates are used to estimate k
cat
values. Another solution is
to infer the apparent k
cat
value (k
app
) in vivo, using measured pro-
teomics and transcriptomics data [77,79].
A second issue is the aforementioned complexity of mammalian
biology compared to simpler systems for which RAMs are more
developed. There still exists a knowledge gap for protein sequences
and gene-protein-reactions associations in mammalian cells, pre-
venting the construction of effective transcription/translation ma-
chinery and integration into metabolism. This could be overcome by
considering a reduced system, for example central carbon metabo-
lism, for which biological understanding is more complete. This can
then be expanded to consider peripheral pathways when the re-
quired data becomes available.
A third issue is the computational burden of fine-grained RAMs.
As an example, one of the original E.coli RAMs [80], contains around
80,000 reactions from an original GEM of around 2000 reactions.
Applying this 40-fold change to Recon 2.2 [55], one of the latest
human GEMs, would result in a model of around 300,000 reactions.
This makes simulation more computationally expensive, which is
particularly problematic for sampling-based approaches. Again, fo-
cusing on a reduced system would alleviate this computational
burden. Overcoming these challenges is imperative to progress
mammalian cell metabolic modelling and to access the benefits that
RAMs can offer to the community.
4. Concluding remarks
Herein, we summarised the main challenges for applying GEMs
and related methodologies to mammalian cell systems, including
human cell systems representative of health and disease states.
These centred around (a) model size, which makes it cumbersome to
apply advanced methodologies and algorithms developed for mi-
crobial cell systems in the absence of significant computational
power, (b) time course data availability, which may be limited to in
vitro studies to avoid intrusive sampling, and (c) the choice of ap-
propriate objective functions that are representative of highly spe-
cialised human cells. Potential solutions involve (a) the integration
Table 1
Resource allocation models and the current challenges in their application to mammalian systems.
Method Method Class Description Current challenges for application to mammalian systems
FBAwMC [57] ecFBA Global constraint on enzyme solvency capacity and
kinetics
Achieved already [76]
MOMENT [81] ecFBA Inclusion of enzyme concentration in solvency
capacity and kinetics
More accurate k
cat
values, further genome annotation
GECKO [60] ecFBA Kinetic and solvency capacity of enzymes with
integration of proteomic data
More accurate k
cat
values, further genome annotation, quantitative
proteomic data
RBA [82] RAM Inclusion and constraining of translation, replication
and transcription machinery
Accurate parameterisation
CAFBA [83] RAM Global constraint modelling tradeoff between growth
and biosynthetic cost
Accurate parameterisation
ME models [53,84] RAM Addition and coupling of transcription and translation
with metabolism
Further genome annotation, quantitative proteomic data, knowledge of
expression machinery, computational burden
ETFL [85] RAM Integration of expression machinery with
thermodynamics
Further genome annotation, quantitative proteomic data, knowledge of
expression machinery, computational burden
B. Strain, J. Morrissey, A. Antonakoudis et al. Computational and Structural Biotechnology Journal 21 (2023) 1543–1549
1547
of data-driven elements with GEMs, either to derive appropriate
constraints that restrict the solution space or to analyse and visualise
GEM results, and (b) the development of RAMs for mammalian and,
eventually, human cell systems. The flexibility that RAMs offer
means that models are widely applicable, beyond exponential cell
growth, where traditional metabolic modelling approaches are less
effective. The main factors restricting mammalian RAM develop-
ment include lack of data and, again, computational burden for large
models. However, it is possible to make small steps towards the goal
of creating full-scale mammalian cell RAMs using microorganism
models as inspiration.
CRediT authorship contribution statement
BS: Conceptualization, Investigation, Visualization, Writing –
original draft. JM: Conceptualization, Investigation, Visualization,
Writing – original draft. AA: Conceptualization, Investigation,
Visualization, Writing – original draft. CK: Conceptualization,
Investigation, Supervision, Writing – review & editing.
Conflict of interest
The authors have no conflict of interest to declare.
Acknowledgements
Benjamin Strain would like to thank the UK Biotechnology and
Biological Sciences Research Council (BBSRC) and GlaxoSmithKline
for their funding and support. James Morrissey thanks the BBSRC
and AstraZeneca for their funding and support. Athanasios
Antonakoudis thanks the UK Engineering and Physical Sciences
Research Council (EPSRC) for their funding and support.
References
[1] Nielsen J. Systems biology of metabolism. Annu Rev Biochem
2017;86(1):245–75.
[2] Maranas C, Zomorrodi A. Flux balance analysis and LP problems. Optimization
methods in metabolic networks. 2016. p. 53–80.
[3] Di Filippo M, Damiani C, Pescini D, GPRuler. Metabolic gene-protein-reaction
rules automatic reconstruction. PLoS Comput Biol 2021;17(11):e1009550.
[4] Haggart CR, et al. Whole-genome metabolic network reconstruction and con-
straint-based modeling. Methods Enzymol 2011;500:411–33.
[5] Martínez VS, et al. The topology of genome-scale metabolic reconstructions
unravels independent modules and high network flexibility. PLoS Comput Biol
2022;18(6):e1010203.
[6] Gu C, et al. Current status and applications of genome-scale metabolic models.
Genome Biol 2019;20(1):121.
[7] Edwards JS, Palsson BO. The Escherichia coli MG1655 in silico metabolic genotype:
its definition, characteristics, and capabilities. Proc Natl Acad Sci USA
2000;97(10):5528–33.
[8] Sheikh K, Förster J, Nielsen LK. Modeling hybridoma cell metabolism using a
generic genome-scale metabolic model of Mus musculus. Biotechnol Prog
2005;21(1):112–21.
[9] Khodaee S, et al. iMM1865: a new reconstruction of mouse genome-scale me-
tabolic model. Sci Rep 2020;10(1):6177.
[10] Tomàs-Gamisans M, Ferrer P, Albiol J. Fine-tuning the P. pastoris iMT1026
genome-scale metabolic model for improved prediction of growth on methanol
or glycerol as sole carbon sources. Microb Biotechnol 2018;11(1):224–37.
[11] Förster J, et al. Genome-scale reconstruction of the Saccharomyces cerevisiae
metabolic network. Genome Res 2003;13(2):244–53.
[12] Duarte NC, et al. Global reconstruction of the human metabolic network based
on genomic and bibliomic data. Proc Natl Acad Sci USA 2007;104(6):1777–82.
[13] Quek L-E, et al. Reducing Recon 2 for steady-state flux analysis of HEK cell
culture. J Biotechnol 2014;184:172–8.
[14] Zhang C, et al. Elucidating the reprograming of colorectal cancer metabolism
using genome-scale metabolic modeling. Front Oncol 2019;9.
[15] Antonakoudis A, et al. The era of big data: genome-scale modelling meets ma-
chine learning. Comput Struct Biotechnol J 2020;18:3287–300.
[16] Kol S, et al. Multiplex secretome engineering enhances recombinant protein
production and purity. Nat Commun 2020;11(1):1908.
[17] Schinn S-M, et al. A genome-scale metabolic network model and machine
learning predict amino acid concentrations in Chinese Hamster Ovary cell cul-
tures. Biotechnol Bioeng 2021;118(5):2118–23.
[18] Antonakoudis A, et al. Synergising stoichiometric modelling with artificial neural
networks to predict antibody glycosylation patterns in Chinese hamster ovary
cells. Comput Chem Eng 2021;154:107471.
[19] Weston BR, Thiele I. A nutrition algorithm to optimize feed and medium com-
position using genome-scale metabolic models. Metab Eng 2023.
[20] Becker SA, Palsson BO. Context-specific metabolic networks are consistent with
experiments. PLoS Comput Biol 2008;4(5):e1000082.
[21] Robaina Estévez S, Nikoloski Z. Generalized framework for context-specific
metabolic model extraction methods. Front Plant Sci 2014;5:491.
[22] Uhlen M, et al. A pathology atlas of the human cancer transcriptome. Science
2017;357(6352):eaan2507.
[23] Paul A, et al. Exploring gene knockout strategies to identify potential drug tar-
gets using genome-scale metabolic models. Sci Rep 2021;11(1):213.
[24] Kishk A, Pacheco MP, Sauter T. DCcov: repositioning of drugs and drug combi-
nations for SARS-CoV-2 infected lung through constraint-based modeling.
iScience 2021;24(11):103331.
[25] Klitgord N, Segrè D. The importance of compartmentalization in metabolic flux
models: yeast as an ecosystem of organelles. Genome Inform 2009:41–55.
[26] Thiele I, Palsson B. A protocol for generating a high-quality genome-scale me-
tabolic reconstruction. Nat Protoc 2010;5(1):93–121.
[27] Fritzemeier CJ, et al. Erroneous energy-generating cycles in published genome
scale metabolic networks: identification and removal. PLoS Comput Biol
2017;13(4):e1005494.
[28] Orth JD, Palsson BØ. Systematizing the generation of missing metabolic
knowledge. Biotechnol Bioeng 2010;107(3):403–12.
[29] Pan S, Reed JL. Advances in gap-filling genome-scale metabolic models and
model-driven experiments lead to novel metabolic discoveries. Curr Opin
Biotechnol 2018;51:103–8.
[30] Wang H, et al. Genome-scale metabolic network reconstruction of model animals
as a platform for translational research. Proc Natl Acad Sci USA 2021;118:30.
[31] Kanehisa M, Goto S. KEGG: kyoto encyclopedia of genes and genomes. Nucleic
Acids Res 2000;28(1):27–30.
[32] Robinson JL, et al. An atlas of human metabolism. Sci Signal 2020;13:624.
[33] Daneshian M, et al. Animal use for science in Europe. Altex 2015;32(4):261–74.
[34] Széliová D, et al. Error propagation in constraint-based modeling of Chinese
hamster ovary cells. Biotechnol J 2021;16(4):2000320.
[35] Széliová D, et al. What CHO is made of: variations in the biomass composition of
Chinese hamster ovary cell lines. Metab Eng 2020;61:288–300.
[36] Noronha A, et al. The Virtual Metabolic Human database: integrating human and
gut microbiome metabolism with nutrition and disease. Nucleic Acids Res
2018;47(D1):D614–24.
[37] Poupin N, et al. Arterio-venous metabolomics exploration reveals major changes
across liver and intestine in the obese Yucatan minipig. Sci Rep 2019;9(1):12527.
[38] Feist AM, Palsson BO. The biomass objective function. Curr Opin Microbiol
2010;13(3):344–9.
[39] Chen Y, et al. An unconventional uptake rate objective function approach en-
hances applicability of genome-scale models for mammalian cells. NPJ Syst Biol
Appl 2019;5:25.
[40] Richelle A, et al. Increasing consensus of context-specific metabolic models by
integrating data-inferred cell functions. PLoS Comput Biol 2019;15(4):e1006867.
[41] Moscardó García M, et al. Importance of the biomass formulation for cancer
metabolic modeling and drug prediction. iScience 2021;24(10):103110.
[42] Lachance J-C, et al. BOFdat: generating biomass objective functions for genome-
scale metabolic models from experimental data. PLoS Comput Biol
2019;15(4):e1006971.
[43] Jones DT. Setting the standards for machine learning in biology. Nat Rev Mol Cell
Biol 2019;20(11):659–60.
[44] Zampieri G, et al. Machine and deep learning meet genome-scale metabolic
modeling. PLoS Comput Biol 2019;15(7):e1007084.
[45] Kim Y, Kim GB, Lee SY. Machine learning applications in genome-scale metabolic
modeling. Curr Opin Syst Biol 2021;25:42–9.
[46] Ramos JRC, et al. Genome-scale modeling of Chinese hamster ovary cells by
hybrid semi-parametric flux balance analysis. Bioprocess Biosyst Eng
2022;45(11):1889–904.
[47] Yaneske E, Angione C. The poly-omics of ageing through individual-based me-
tabolic modelling. BMC Bioinform 2018;19(14):415.
[48] Lewis JE, Kemp ML. Integration of machine learning and genome-scale metabolic
modeling identifies multi-omics biomarkers for radiation resistance. Nat
Commun 2021;12(1):2700.
[49] De Becker K, et al. Using resource constraints derived from genomic and pro-
teomic data in metabolic network models. Curr Opin Syst Biol 2022;29:100400.
[50] Kerkhoven EJ. Advances in constraint-based models: methods for improved
predictive power based on resource allocation constraints. Curr Opin Microbiol
2022;68:102168.
[51] Chen Y, Nielsen J. Mathematical modeling of proteome constraints within me-
tabolism. Curr Opin Syst Biol 2021;25:50–6.
[52] Dahal S, Zhao J, Yang L. Recent advances in genome-scale modeling of proteome
allocation. Curr Opin Syst Biol 2021;26:39–45.
[53] O'Brien EJ, et al. Genome-scale models of metabolism and gene expression ex-
tend and refine growth phenotype prediction. Mol Syst Biol 2013;9:693.
[54] Massaiu I, et al. Integration of enzymatic data in Bacillus subtilis genome-scale
metabolic model improves phenotype predictions and enables in silico design of
poly-γ-glutamic acid production strains. Microb Cell Fact 2019;18(1):3.
[55] Swainston N, et al. Recon 2.2: from reconstruction to model of human meta-
bolism. Metabolomics 2016;12:109.
B. Strain, J. Morrissey, A. Antonakoudis et al. Computational and Structural Biotechnology Journal 21 (2023) 1543–1549
1548
[56] Hefzi H, et al. A consensus genome-scale reconstruction of Chinese hamster
ovary cell metabolism. Cell Syst 2016;3(5):434–43. e8.
[57] Beg QK, et al. Intracellular crowding defines the mode and sequence of substrate
uptake by Escherichia coli and constrains its metabolic activity. Proc Natl Acad Sci
USA 2007;104(31):12663–8.
[58] Molenaar D, et al. Shifts in growth strategies reflect tradeoffs in cellular eco-
nomics. Mol Syst Biol 2009;5:323.
[59] Zhuang K, Vemuri GN, Mahadevan R. Economics of membrane occupancy and
respiro-fermentation. Mol Syst Biol 2011;7:500.
[60] Sánchez BJ, et al. Improving the phenotype predictions of a yeast genome-scale
metabolic model by incorporating enzymatic constraints. Mol Syst Biol
2017;13(8):935.
[61] Scott M, et al. Interdependence of cell growth and gene expression: origins and
consequences. Science 2010;330(6007):1099–102.
[62] Basan M, et al. Overflow metabolism in Escherichia coli results from efficient
proteome allocation. Nature 2015;528(7580):99–104.
[63] Han X, et al. Cancer causes metabolic perturbations associated with reduced
insulin-stimulated glucose uptake in peripheral tissues and impaired muscle
microvascular perfusion. Metabolism 2020;105:154169.
[64] Läsche M, Emons G, Gründker C. Shedding new light on cancer metabolism: a
metabolic tightrope between life and death. Front Oncol 2020:10.
[65] Vanhove K, et al. The metabolic landscape of lung cancer: new insights in a
disturbed glucose metabolism. Front Oncol 2019:9.
[66] Stine ZE, et al. Targeting cancer metabolism in the era of precision oncology. Nat
Rev Drug Discov 2022;21(2):141–62.
[67] Zagari F, et al. Lactate metabolism shift in CHO cell culture: the role of mi-
tochondrial oxidative activity. New Biotechnol 2013;30(2):238–45.
[68] Brunner M, et al. Elevated pCO(2) affects the lactate metabolic shift in CHO cell
culture processes. Eng Life Sci 2018;18(3):204–14.
[69] Ahleboot Z, et al. Designing a strategy for pH control to improve CHO cell pro-
ductivity in bioreactor. Avicenna J Med Biotechnol 2021;13(3):123–30.
[70] Chen Y, et al. Proteome constraints reveal targets for improving microbial fitness
in nutrient-rich environments. Mol Syst Biol 2021;17(4):e10093.
[71] Kim MK, et al. E-Flux2 and SPOT: validated methods for inferring intracellular
metabolic flux distributions from transcriptomic data. PLoS One
2016;11(6):e0157101.
[72] Zur H, Ruppin E, Shlomi T. iMAT: an integrative metabolic analysis tool.
Bioinformatics 2010;26(24):3140–2.
[73] Jensen PA, Papin JA. Functional integration of a metabolic network model and
expression data without arbitrary thresholding. Bioinformatics
2011;27(4):541–7.
[74] Lerman JA, et al. In silico method for modelling metabolism and gene product
expression at genome scale. Nat Commun 2012;3(1):929.
[75] Lloyd CJ, et al. Computation of condition-dependent proteome allocation reveals
variability in the macro and micro nutrient requirements for growth. PLoS
Comput Biol 2021;17(6):e1007817.
[76] Yeo HC, et al. Enzyme capacity-based genome scale modelling of CHO cells.
Metab Eng 2020;60:138–47.
[77] Davidi D, et al. Global characterization of in vivo enzyme catalytic rates and their
correspondence to in vitro kcat measurements. Proc Natl Acad Sci USA
2016;113(12):3401–6.
[78] Li F, et al. Deep learning-based kcat prediction enables improved enzyme-con-
strained model reconstruction. Nat Catal 2022;5(8):662–72.
[79] Heckmann D, et al. Kinetic profiling of metabolic specialists demonstrates sta-
bility and consistency of in vivo enzyme turnover numbers. Proc Natl Acad Sci
USA 2020;117(37):23182–90.
[80] Thiele I, et al. Multiscale modeling of metabolism and macromolecular synthesis
in E. coli and its application to the evolution of codon usage. PLoS One
2012;7(9):e45635.
[81] Adadi R, et al. Prediction of microbial growth rate versus biomass yield by a
metabolic network with kinetic parameters. PLoS Comput Biol
2012;8(7):e1002575.
[82] Goelzer A, Fromion V, Scorletti G. Cell design in bacteria as a convex optimiza-
tion problem. Automatica 2011;47(6):1210–8.
[83] Mori M, et al. Constrained allocation flux balance analysis. PLoS Comput Biol
2016;12(6):e1004913.
[84] Thiele I, et al. Genome-scale reconstruction of Escherichia coli's transcriptional
and translational machinery: a knowledge base, its mathematical formulation,
and its functional characterization. PLoS Comput Biol 2009;5(3):e1000312.
[85] Salvy P, Hatzimanikatis V. The ETFL formulation allows multi-omics integration
in thermodynamics-compliant metabolism and expression models. Nat
Commun 2020;11(1):30.
B. Strain, J. Morrissey, A. Antonakoudis et al. Computational and Structural Biotechnology Journal 21 (2023) 1543–1549
1549
