Table 1 Summary of developed integration algorithms to generate context-specific GEMs.
Algorithm | Year | Organism | Description | Programming language |
|---|---|---|---|---|
Covert-013 | 2001 | — | Introduction of a theoretical framework to broaden the predictive capabilities of GEMs with the incorporation of a transcriptional regulatory network using Boolean logic formalism. | NA |
Covert-0226 | 2002 | E. coli | The previous framework, applied to the central metabolism in E.coli. | NA |
Akesson-0427 | 2004 | S. cerevisiae | Introduction of a theoretical framework for the integration of transcriptome data into metabolic models. In this framework, the flux of some reactions constraint to zero, if their associated genes are not expressed. | NA |
SR‐FBA28 | 2007 | E. coli | SR-FBA uses the same framework as Covert-01, but with a different formulation. SR-FBA identifies genes and metabolic functions in which regulation is not optimally tuned for cellular flux demands. | NA |
Shlomi-0817 | 2008 | H. sapiens, and S. cerevisiae | The integrative metabolic analysis (iMAT) framework modified the Boolean mapping for tri-valued gene expression levels; high, low, and moderately expressed genes. | NA |
GIMME9 | 2008 | E. coli, and H. sapiens | GIMME presented a conceptual framework that indicates how consistent a set of gene expression data is with a desired metabolic objective. | MATLAB + COBRA |
E-Flux29 | 2009 | M. tuberculosis | In contrast to previously developed methods for metabolically interpreting gene expression data, E-Flux uses the normalized gene expression level to approximate and constrain the maximum flux of corresponding reactions. | MATLAB + COBRA |
Moxley30 | 2009 | S. cerevisiae | In this framework, the developers considered the correlation between changes in gene expression and reaction levels as well as topological parameters of the metabolic networks. | NA |
MBA31 | 2010 | H. sapiens | This framework has been developed to integrate a variety of omics data as well as literature‐based knowledge to generate context-specific models. Accordingly, the algorithm takes two reaction sets as input (extracted from omics data analysis and literature) and reconstructs a model containing as many as possible of input reactions, and a minimal set of reactions that are required for model consistency. | MATLAB + COBRA |
MADE32 | 2011 | S. cerevisiae | Selecting arbitrary thresholds to find active/deactive reactions was the main challenge in previously developed algorithms. MADE tackles this problem by comparing gene expression levels across multiple conditions to identify activation/deactivation patterns. | MATLAB + COBRA |
tFBA33 | 2011 | S. cerevisiae | tFBA uses the same framework as MADE, but with a different formulation. Basically, tFBA assumes that if the activity of a gene in two different conditions drastically changes, the associated flux will change accordingly. | NA |
RELATCH34 | 2012 | E. coli, S. cerevisiae, and B. subtilis | RELATCH defined a “relative optimality” conceptual framework to create a context-specific model. In this framework, perturbed cells preserve a relative metabolic flux pattern from a reference state using metabolic adaptation and regulatory reprogramming. | MATLAB + COBRA |
INIT16 | 2012 | H. sapiens | INIT integrates proteomic or transcriptomic data into a GEM. The INIT algorithm maximizes the activation of certain reactions associated with highly expressed genes while minimizing the utilization of reactions associated with absent proteins. | MATLAB + COBRA + RAVEN |
mCADRE35 | 2012 | H. sapiens | mCARDE uses the same framework as MBA, but with a different formulation for non-core reactions. mCARDE ranks non-core reactions according to their expression as well as weighted connectivity in the given network, and then keeps reactions which are required for model consistency. | MATLAB + COBRA |
AdaM36 | 2012 | E. coli | AdaM provides a framework to evaluate adaptation upon external perturbation. AdaM extracts minimal operating networks (to characterize the transitional behavior) by integration of time-series transcriptomics data with flux-based bi-level optimization. | NA |
Lee-1237 | 2012 | S. cerevisiae | In this theoretical framework, the objective function of a context-specific model is justified by gene expression data, while the flux constraints of the model remain untouched. | MATLAB + COBRA |
Fang-1238 | 2012 | M. tuberculosis | This approach works according to a) relative gene expression between a reference state and a perturbed condition, b) the correlation between gene transcription levels and enzymatic activity, c) the justification of the biomass composition for the perturbed state. | MATLAB + COBRA |
GX–FBA39 | 2012 | Y. pestis, and S. cerevisiae | GX-FBA optimizes the pattern of hierarchical regulation, as well as the level of differential gene-expression within the framework of metabolic constraints. | MATLAB + COBRA |
TEAM40 | 2012 | S. oneidensis | TEAM uses the same framework as GIMME, but differs in that it takes advantage of large compendium of gene expression data to estimate each gene’s unique transcriptional signature. | NA |
GIM3E41 | 2013 | S. typhimurium | GIM3E uses the same framework as GIMME, but in this algorithm, the estimated penalties minimize reactions that have weaker supporting evidence. | Python+COBRApy |
EXAMO42 | 2013 | S. cerevisiae | EXAMO uses the same framework as iMAT, but with a different formulation. EXAMO considers high-frequency reactions (HFR) as active reactions and then minimizes the given GEM so that all HFR should be able to carry flux. | Python standalone |
MTA43 | 2013 | E. coli, S. cerevisiae, M. musculus, and H. sapiens | MTA uses the same integration framework as iMAT, with different application for drug target prediction. In fact, MTA identifies drug targets that alter the metabolism in order to retrieve it back to the initial healthy state. | NA |
FASTCORE44 | 2014 | H. sapiens | FASTCORE uses the same framework as MBA, but with a different formulation. Accordingly, FASTCORE takes a core set of active reactions as input, and then searches for a flux consistent subnetwork. | MATLAB + COBRA |
tINIT45 | 2014 | H. sapiens | INIT-based GEMs are not functional models and could not be used for simulations. Therefore, tINIT developers tackled this issue by defining metabolic tasks, which the resulting model should be able to perform. | NA |
E-Fmin46 | 2014 | E. coli, and S. cerevisiae | E-Fmin is a modified extension of GIMME which forces biomass production to carry non-zero flux, whereas GIMME requires certain metabolic functionalities to be active above condition-dependent thresholds. | MATLAB |
METRADE47 | 2015 | E. coli | METRADE developed a multi-omic framework that integrates gene expression and codon usage into the GEM, and uses a multi-objective optimization algorithm to optimize metabolic phenotypes. | MATLAB |
Lsei-FBA48 | 2015 | H. sapiens | Lsei-FBA’s developers argued that the performance of existing methods was optimized for large changes in gene expression. To tackle this issue, Lsei-FBA uses the fold changes in mRNA gene expression to estimate the changes in the metabolic network. | R-package |
FASTCORMICS49 | 2015 | H. sapiens | FASTCORMICS is adapted FASTCORE for the direct integration of microarray data. | MATLAB + COBRA |
TREM-Flux50 | 2015 | C. reinhardtii | This algorithm integrates time-resolved metabolomics and transcriptomics data (performed based on E-Flux method). | MATLAB |
RegrEx51 | 2015 | H. sapiens | RegrEx is a fully automated framework that extracts a context-specific model by maximizing the correlation of flux distribution and a given data. | MATLAB + COBRA |
CORDA52 | 2016 | H. sapiens | CORDA is based on a dependency assessment approach which identifies the dependency of desirable reactions (i.e. reactions with high experimental evidence) on undesirable reactions (i.e. reactions with no experimental evidence). | MATLAB + COBRA |
OM-FBA53 | 2016 | S. cerevisiae | OM-FBA uses the same framework as Lee-12, but it deploys a “Phenotype Match” algorithm to derive an objective function to be correlated with the transcriptomics data via regression analysis. | MATLAB |
E-Flux254 | 2016 | E. coli, and S. cerevisiae | E-Flux2 is an extension of E-Flux which is combined with minimization of l2 norm. | MATLAB+Java+ (MOST) |
SPOT54 | 2016 | E. coli, and S. cerevisiae | SPOT uses pearson correlation with transcriptomic data that can be combined with carbon source or objective function. | MATLAB+Java+ (MOST) |
metaboGSE55 | 2018 | Y. lipolytica, and M. musculus | metaboGSE introduced a new framework of creating condition-specific models. This algorithm creates a series of metabolic sub-networks by removing genes from a GEM. Then, sub-networks are evaluated via a fitness function. | R-package |
Benchmark-driven56 | 2019 | H. sapiens | Benchmark-driven approach is the last developed framework that is devised based on the advantageous features and bottlenecks of the selected algorithms. | MATLAB |
