Skip to content

Latest commit

 

History

History
705 lines (540 loc) · 32.1 KB

File metadata and controls

705 lines (540 loc) · 32.1 KB

COBRA Toolbox vs RAVEN Toolbox — Functionality Comparison

This document compares the COBRA Toolbox (MATLAB) and RAVEN Toolbox (MATLAB) feature-by-feature. The primary aim is to identify what COBRA has that RAVEN lacks, with particular attention to genome-scale model reconstruction — RAVEN's core purpose — but covering all functional areas.

Sources

  • COBRA Toolbox: v3.x documentation, module index at opencobra.github.io, and the COBRA v3.0 paper (arXiv:1710.04038)
  • RAVEN Toolbox: RAVEN v2.x source tree (analysis/, reconstruction/, io/, external/, core/)

Contents

  1. Executive summary
  2. Genome-scale model reconstruction
  3. Model editing and manipulation
  4. Flux balance analysis and LP/MILP solving
  5. Flux variability and enumerating alternatives
  6. Knockout and deletion analysis
  7. Strain design and metabolic engineering
  8. Omics-driven context-specific modelling
  9. Thermodynamics
  10. Flux sampling
  11. Multi-species and microbiome modelling
  12. Network topology and elementary flux modes
  13. Visualization and reporting
  14. Metabolomics integration
  15. Model I/O and database connectivity
  16. Summary gap table
  17. RAVEN strengths not in COBRA

1. Executive summary

RAVEN and COBRA serve overlapping but distinct purposes.

RAVEN's core strength is automated genome-scale model reconstruction: pulling protein sequences from a genome, running HMM-based homology search against curated databases, scoring gene-reaction associations, and extracting tissue-specific models via tINIT/ftINIT from transcriptomic data. Its analysis toolbox covers the essential FBA operations but is narrower than COBRA's.

COBRA's core strength is constraint-based analysis breadth: thermodynamic integration, advanced strain design (OptKnock, OptForce, gMCS), multiple loopless and parsimonious FBA variants, Markov-chain flux sampling, microbiome community models, and whole-body physiology models. Its reconstruction pipeline (DEMETER, rBioNet) targets microbial and human databases and is less focused on de-novo genome-based reconstruction.

For a lab doing human tissue or yeast GEM reconstruction from genomic or transcriptomic data, RAVEN covers the critical path well. For labs doing thermodynamically constrained analysis, strain design, microbiome modelling, or multi-omics integration beyond transcriptomics, COBRA has capabilities that RAVEN lacks entirely.


2. Genome-scale model reconstruction

RAVEN

Function Description
getKEGGModelForOrganism Full KEGG-based draft reconstruction from organism taxonomy ID; pulls protein sequences, runs DIAMOND BLASTP against KEGG protein database, scores gene-reaction associations
getModelFromHomology Template-based reconstruction via bidirectional HMM/BLAST homology; the backbone of high-quality GEM reconstruction from reference models
makeRaven Assembles a RAVEN model struct from scored homology results
addRxnsGenesMets Adds reactions, genes, and metabolites to an existing draft
gapReport Reports metabolic gaps (dead ends, disconnected subsystems)
fillGaps LP-based heuristic gap-filling against a universal reaction database
runINIT INIT (Integrative Network Inference for Tissues) — tissue-specific draft from expression scores
tINIT The original INIT implementation
ftINIT Fast tINIT with MILP-based extraction; state-of-the-art for human tissue models
getINITModel2 Wrapper calling tINIT/ftINIT with HPA/GTEx data loading
checkModel Validates RAVEN model struct fields
simplifyModel Removes empty genes/metabolites, normalises identifiers

COBRA

Function Description
fastGapFill MILP-based gap-filler (Thiele et al. 2014); fills stoichiometric gaps using a universal reaction database (KEGG + VMH) including transport reactions
swiftGapFill Alternative MILP gap-filler using SWIFTCORE
prepareFastGapFill Builds the S/U/X matrix input for fastGapFill
generateSUXMatrix Assembles combined stoichiometric + universal + transport matrices
gapFind Identifies blocked metabolites (anything downstream of a gap)
identifyBlockedRxns Detects all blocked reactions via FASTCORE
postProcessGapFillSolutions Classifies added reactions (metabolic/transport/exchange)
fastcc Tests stoichiometric consistency; returns the flux-consistent subset
checkCobraModelUnique Checks uniqueness of reaction/metabolite names
checkDuplicateRxn Detects and removes duplicate reactions
checkModelPreFBA Validates stoichiometric and flux consistency before FBA
detectDeadEnds Returns metabolites participating in only one reaction
fastLeakTest Tests whether metabolites leak in a closed model
removeDeadEnds Removes dead-end metabolites and their reactions
DEMETER pipeline Automated refinement of microbial draft reconstructions from KBase/ModelSEED; includes taxonomy-based GPR refinement and 100+ quality tests
rBioNet GUI-based reconstruction tool with VMH database integration
Model borgifier Submodule for merging and comparing two reconstructions

Gaps (COBRA has, RAVEN lacks)

  • MILP gap-filling: RAVEN's fillGaps is an LP heuristic; COBRA's fastGapFill / swiftGapFill use MILP to find the globally minimal set of reactions. This produces more parsimonious fills.
  • Stoichiometric consistency checking: fastcc, fastLeakTest, checkDuplicateRxn, detectDeadEnds. RAVEN's checkModel validates struct fields but does not test flux consistency.
  • DEMETER / KBase pipeline: For building or refining microbial community member models from KBase/ModelSEED outputs, COBRA has a dedicated automated pipeline.
  • VMH / BiGG-centred reconstruction: If building models against the VMH or BIGG namespace, COBRA's rBioNet and DEMETER pipelines are tailored for it.

RAVEN strengths not in COBRA

  • KEGG-centric genome reconstruction: getKEGGModelForOrganism is a mature, maintained pipeline that maps a genome directly to KEGG reactions and produces a draft requiring only gap-filling. COBRA has no equivalent end-to-end KEGG-based reconstruction.
  • HMM-based homology scoring: getModelFromHomology supports HMMER as the search engine, giving finer granularity on gene confidence than BLAST alone.
  • ftINIT quality: For human tissue-specific models based on HPA/GTEx protein/RNA data, ftINIT (RAVEN) is the state-of-the-art; COBRA's equivalent methods (GIMME, iMAT, FASTCORE) predate it and are generally less accurate on human models.

3. Model editing and manipulation

RAVEN

RAVEN has a comprehensive set of model manipulation utilities:

addRxns, addMets, addGenes, removeRxns, removeMets, removeGenes, setParam (constraint modification), setExchangeBounds, getExchangeRxns, contractModel (merge identical reactions), expandModel (split reversible reactions), convertToReversible, convertToIrreversible, mergeModels, combineModels, getModelDiff, sortModel, copyToRaven (struct copy), exportForGit (deterministic field ordering for version control).

COBRA

addReaction, removeRxns, addMetabolite, changeRxnBounds, changeObjective, convertToIrreversible, mergeModels, findExcRxns, extractSubNetwork, subNetwork, buildRxnGeneMat, tncore (gene essentiality-driven network reduction), pruneModel (removes blocked reactions below a flux threshold).

Assessment

Both toolboxes cover the essentials. RAVEN's exportForGit and getModelDiff are distinctive for version-controlled model development. COBRA's tncore and extractSubNetwork are useful for focused subnetwork analysis.


4. Flux balance analysis and LP/MILP solving

RAVEN

Function Description
solveLP Standard LP: maximise/minimise c'v subject to Sv=0, lb≤v≤ub
optimizeProb General LP/MILP solver wrapping GLPK, Gurobi, CPLEX, MOSEK
getMinNrFluxes MILP: find flux distribution with minimum number of active reactions
getAllowedBounds Flux variability: min/max flux for each reaction
haveFlux Identifies reactions that can carry non-zero flux
getEssentialRxns Returns reactions whose deletion abolishes objective
getMinimalMedium MILP: minimal set of uptake exchanges that allow growth
runDynamicFBA Dynamic FBA (Euler method, static optimisation)

COBRA

Function Description
optimizeCbModel Flexible FBA: LP, MILP, QP, zero-norm cardinality
pFBA Parsimonious FBA (Lewis et al. 2010): optimise objective, then minimise total flux; classifies gene usage
geometricFBA Finds unique central point inside optimal FBA polytope (Smallbone & Simeonidis 2009)
sparseFBA Finds minimal set of active reactions achieving objective (L0)
relaxedFBA Finds minimal relaxations of bounds to make infeasible model feasible
relaxFBA_cappedL1 Same via capped-L1 norm approximation
enumerateOptimalSolutions Finds multiple alternative optimal flux distributions
minimizeModelFlux Minimises total flux (taxicab, zero-norm, or Euclidean)
cycleFreeFlux Removes stoichiometrically balanced cycles from FBA solutions
addLoopLawConstraints Adds thermodynamic loop-law constraints to LP/MILP
dynamicFBA Dynamic FBA with substrate/biomass dynamics (Varma & Palsson 1994)
dynamicRFBA Dynamic regulatory FBA; time evolution with Boolean regulatory network
mdFBA Metabolite dilution FBA — accounts for growth-rate-dependent metabolite dilution

Gaps

  • pFBA: RAVEN has no parsimonious FBA. Gene classification (essential / pFBA-optimal / enzymatically complex / metabolically blocked) requires COBRA.
  • Loopless FBA: RAVEN has no cycle removal or loop-law constraints. Solutions can contain thermodynamically infeasible cycles; COBRA's cycleFreeFlux / addLoopLawConstraints remove them.
  • Geometric FBA: No equivalent in RAVEN.
  • Infeasibility relaxation: RAVEN returns infeasibility; relaxedFBA finds the minimal bound adjustments needed to restore feasibility.
  • Multiple optimal solutions: RAVEN has no equivalent to enumerateOptimalSolutions.
  • Regulatory dynamic FBA: dynamicRFBA integrates a Boolean gene-regulatory network over time; RAVEN has no equivalent.

5. Flux variability and enumerating alternatives

RAVEN

getAllowedBounds — basic FVA for a subset of reactions. Single-threaded.

COBRA

Function Description
fluxVariability Full FVA; supports loopless mode, multi-threading, and per-reaction objective constraints
fastFVA Accelerated FVA via multi-threaded CPLEX solver
mtFVA Multi-threaded FVA
fvaJaccardIndex Compares flux ranges between two models using Jaccard index

Gaps

  • Loopless FVA mode — eliminates cycle-containing solutions from variability ranges.
  • Multi-threaded FVA — faster for large models.
  • Model-to-model FVA comparison (fvaJaccardIndex).

6. Knockout and deletion analysis

RAVEN

Function Description
findGeneDeletions Single gene deletion via FBA, returns growth for each KO

COBRA

Function Description
singleGeneDeletion Single gene KO via FBA, MOMA, or linearMOMA
singleRxnDeletion Single reaction KO
doubleGeneDeletion All pairwise gene knockout combinations
MOMA Quadratic MOMA — minimises flux adjustment between wild-type and KO
linearMOMA Linear MOMA variant (LP)
ROOM Regulatory on/off minimisation (MILP) — minimises number of altered reactions
linearROOM LP relaxation of ROOM
findEpistaticInteractions Double deletion analysis for synthetic sick/lethal pairs
essentialRxn4MultipleModels Reaction deletion across multiple model variants
Fast-SL submodule Identifies all synthetic lethal gene pairs

Gaps

  • Quadratic / ROOM knockout prediction: After a gene deletion, MOMA predicts the most likely adaptive flux redistribution by minimising deviation from wild-type fluxes; ROOM minimises the number of reactions that change. Both are more biologically realistic than simply re-optimising the objective. RAVEN has neither.
  • Double gene deletion: RAVEN only supports single-gene knockouts. COBRA's doubleGeneDeletion runs pairwise combinatorial analysis.
  • Synthetic lethality: RAVEN has no equivalent to Fast-SL.

7. Strain design and metabolic engineering

RAVEN

Function Description
FSEOF Flux Scanning based on Enforced Objective Flux — ranks reactions by co-flux with target
runProductionEnvelope Production envelope for biomass vs. target product
runRobustnessAnalysis Objective sensitivity as a reaction's flux varies
runPhenotypePhasePlane 2-reaction phenotype phase plane
runSimpleOptKnock Simplified single-level OptKnock approximation

COBRA

OptKnock family

Function Description
OptKnock Bilevel MILP to find knockout sets that couple biomass growth with target overproduction
analyzeOptKnock Validates and analyses OptKnock solutions
createBilevelMILPproblem Low-level bilevel MILP builder
optGene Evolutionary programming-based knockout strategy
GDLS Genetic Design by Local Search — fast neighbourhood search

OptForce

Function Description
FVAOptForce FVA for wild-type vs. mutant with production constraints
findMustU / findMustL First-order must-upregulate / must-downregulate sets
findMustUU / findMustUL / findMustLL Second-order must-sets (pairwise)
optForce Finds minimal k-intervention sets for targeted overproduction
analyzeOptForceSol Evaluates proposed interventions

Minimal Cut Sets (drug targets / growth coupling)

Function Description
calculateGeneMCS Computes genetic Minimal Cut Sets (gMCSs) using CPLEX warm-start
calculateMCS Reaction-level minimal cut sets
checkGeneMCS Validates minimality of computed gMCSs

OptEnvelope

Function Description
optEnvelope Finds minimum knockout set giving equivalent production envelope
minActiveRxns / sequentialOEReinserts / milpOEReinserts Supporting envelope optimisation

Other

Function Description
productionEnvelope Richer production envelope with flux scanning variants
theoretMaxProd Theoretical maximum production for a target
phenotypePhasePlane 3D surface + shadow prices for two reactions

Gaps

COBRA's strain design toolbox is far broader than RAVEN's:

  • True OptKnock (bilevel MILP, not heuristic): RAVEN's runSimpleOptKnock uses a single-level approximation; COBRA has the full bilevel formulation.
  • OptForce: Identifies reactions that must change (up or down) to achieve a production target — a complementary perspective to knockout-only methods.
  • GDLS / optGene: Heuristic design strategies for when the MILP becomes intractable.
  • Genetic MCS: calculateGeneMCS computes Minimal Cut Sets at the gene level — essential for identifying combination drug targets or growth-coupled designs.

8. Omics-driven context-specific modelling

RAVEN

Function Description
tINIT / ftINIT INIT and fast-INIT (MILP) for tissue-specific models from HPA/GTEx expression data
getINITModel2 High-level wrapper loading HPA/GTEx data and calling ftINIT
scoreComplexModel Scores gene-reaction-metabolite associations from expression data
parseScores Converts raw expression matrix to per-reaction scores

COBRA

Algorithm suite (all under createTissueSpecificModel)

Algorithm Reference
GIMME Becker & Palsson 2008 — minimises low-expression reactions while maintaining objective
iMAT Zur et al. 2010 — balances inclusion of high- vs. exclusion of low-expression reactions
FASTCORE Vlassis et al. 2014 — extracts minimal model containing a user-defined core
SWIFTCORE Fastest consistency-based extraction
MBA Model Building Algorithm — uses high/medium confidence reaction sets
eFlux Constrains reaction bounds proportional to gene expression
MOOMIN Metabolic model-based optimisation of omics integration

Multi-omics

Function Description
XomicsToModel Integrates transcriptomics + proteomics + metabolomics + literature into a thermodynamically consistent context-specific model
XomicsToMultipleModels Batch generation of model variants with different parameter combinations
modelExtraction Sub-step within XomicsToModel

Gaps

  • Multiple extraction algorithms: RAVEN only implements INIT/ftINIT. For benchmarking or when HPA/GTEx data are unavailable (e.g., microorganisms), COBRA's suite (GIMME, iMAT, FASTCORE, etc.) offers alternatives.
  • Proteomics + metabolomics integration in model extraction: XomicsToModel combines transcriptomics, proteomics, metabolomics, and literature curation in one pipeline. RAVEN's pipeline is transcriptomics-only.
  • Note: For human tissue models where HPA/GTEx data are available, ftINIT is the current gold standard; COBRA's methods are not inherently superior here.

9. Thermodynamics

RAVEN

None. RAVEN has no thermodynamic constraint methods.

COBRA

vonBertalanffy / thermoFBA module

Function Description
setupThermoModel Estimates standard transformed Gibbs energies (ΔrG'°) and sets directional constraints at in vivo conditions; requires ChemAxon/Open Babel
initVonBertalanffy Initialises the module
estimateDfGt0 / estimateDrGt0 Gibbs energies of formation/reaction with uncertainty bounds
estimateDG_temp ΔG at specified temperature
configureSetupThermoModelInputs Sets compartment pH, ionic strength, concentration bounds
cycleFreeFlux Cycle-free (loopless) FBA solution
addLoopLawConstraints Embeds thermodynamic loop constraints into LP/MILP
consistentPotentials Chemical potentials satisfying thermodynamic directionality
checkThermodynamicConsistency Validates thermodynamic feasibility
fastSNP Generates minimal feasible basis for internal cycles (Saa & Nielsen 2016)
findMinNull Minimal nullspace for internal cycles via MILP (Chan et al. 2017)

Gaps

RAVEN has no thermodynamic capabilities at all. If thermodynamic driving force analysis, Max-min Driving Force (MDF), or thermodynamically feasible flux distributions are needed, COBRA (or the separate matTFA toolbox) is required.


10. Flux sampling

RAVEN

Function Description
randomSampling Random-objective flux sampling — samples diverse flux states by solving FBA with random objective vectors
analyzeSampling Correlates per-reaction expression t-scores with flux samples; returns enrichment scores
getFluxZ Computes Z-scores comparing two flux sample sets

COBRA

Function Description
sampleCbModel Unified entry point; dispatches to ACHR, CHRR, or RHMC
ACHR sampler Artificial centering hit-and-run — ergodic, good convergence
CHRR sampler Coordinate hit-and-run with rounding — uniform sampling from flux polytope
RHMC sampler Riemannian manifold Hamiltonian Monte Carlo — fastest for high-dimensional models
calcSampleStats Mean, SD, mode, median, skewness, kurtosis
calcSampleDifference Pairwise differences between two sample sets
compareSampleTraj Compares flux histograms across samples
compareTwoSamplesStat KS test, rank-sum, chi-square, T-test between samples
identifyCorrelSets Finds correlated reaction groups in flux samples
plotHistConv / plotSampleHist / sampleScatterMatrix Visualisation and diagnostics

Gaps

  • Markov-chain sampling (ACHR, CHRR, RHMC): these produce provably uniform samples from the feasible flux polytope. RAVEN's random-objective approach is fast but does not guarantee uniform coverage, and is known to under-sample extreme flux states.
  • Statistical comparison tools: COBRA has built-in tests for comparing two sample sets; RAVEN's getFluxZ is a lightweight scoring function, not a hypothesis test.

11. Multi-species and microbiome modelling

RAVEN

None.

COBRA

SteadyCom (community metabolic models)

Function Description
SteadyCom Finds maximum community growth rate at community steady-state
SteadyComFVA FVA on community models across growth rates
SteadyComPOA Pairwise analysis of member biomasses and reaction fluxes

Microbiome Modeling Toolbox / mgPipe (AGORA2 pipeline)

Function Description
mgPipe / initMgPipe Integrates metagenomic abundances with AGORA member models
fastSetupCreator Builds microbiota community model with lumen/fecal compartments
createPersonalizedModel Individual-specific community models from abundance data
buildModelStorage Adds internal exchange space and coupling constraints
createPanModels Pan-models at species/genus/family level
microbiotaModelSimulator Applies diet constraints, runs FVA
adaptVMHDietToAGORA Converts VMH diet definitions to AGORA-compatible constraints
translateMetagenome2AGORA Maps metagenomic taxonomic IDs to AGORA model IDs

Whole-body physiological models (Harvey/Harvetta, Recon3D-based)

Function Description
optimizeWBModel LP optimisation on whole-body models
Test4HumanFctExtv5 Tests ~460 human metabolic functions
checkIEM_WBM Simulates inborn errors of metabolism in whole-body models
performIEMAnalysis Tests biomarker metabolites in biofluids for genetic disorders
organEssentiality Determines organ essentiality
calcOrganFract / calculateBMR Physiological parameterisation

Gaps

RAVEN has no multi-species or microbiome modelling capability. These are substantial, specialised areas and their absence reflects RAVEN's focus on single-organism reconstruction.


12. Network topology and elementary flux modes

RAVEN

Function Description
getAllSubGraphs Identifies metabolite connectivity subgraphs (connected components)
traceFluxPath Traces the carbon/flux path between two reactions via metabolite producers/consumers
compareFluxes Diffs two flux distributions; reports turned-on, turned-off, flipped, and changed reactions
followFluxes Interactive command-line exploration of a flux distribution
followChanged Displays the reactions that changed most between two conditions

COBRA

Function Description
findBlockedReaction / findBlockedIrrRxns Reactions with zero flux capacity
findMetabolicJunctions Identifies metabolic branchpoints
findCarbonRxns Finds reactions above a carbon count threshold
conservedMoieties Computes conserved moieties (L-matrix)
reactingMoieties Identifies reacting moiety pairs
computeExtremeRays Computes extreme rays of the flux cone
EFMviz suite Wrapper for EFMtool (Java): import, filter, yield analysis, enrichment analysis, backbone extraction, SBML export of EFM submodels
surfNet Interactive network navigation

Gaps

  • Elementary flux modes (EFMs): COBRA interfaces with EFMtool to compute and analyse EFMs. RAVEN has no EFM support.
  • Conserved moieties: Useful for detecting modelling errors and for metabolite grouping. Not present in RAVEN.
  • Extreme ray computation: Not present in RAVEN.

RAVEN strengths

  • traceFluxPath and compareFluxes are practical diagnostic tools not present in COBRA; they make it easy to understand where flux flows and what changed between conditions.

13. Visualization and reporting

RAVEN

Function Description
drawFluxes Network diagram with reaction fluxes overlaid
plotFluxes Bar/scatter plot of flux distributions
printFluxes Formatted text table of a flux solution
followFluxes CLI exploration of flux paths
Various plotting helpers in analysis/ Phase planes, robustness plots, production envelopes

COBRA

Function Description
paint4Net draw_by_met, draw_by_rxn, bio_paint4net — renders metabolic networks
Escher integration Exports model/flux data to Escher JSON for interactive browser maps
CellDesigner conversion Converts models to CellDesigner XML
EFMviz Visualises EFMs as reaction/metabolite/gene networks
ReconMap Interface to VMHDB/ReconMap web cartography
metaboliteMassBalancePlot Bar chart of top producers/consumers of a metabolite
upsetPlot Upset plots for set comparisons across models
visualizeEpistasis Heatmap/network visualisation of double-deletion interactions
plotEssentialRxns Heatmap of essential reactions across models
phenotypePhasePlane 3D surface + shadow prices
plotSampleHist / sampleScatterMatrix Sampling diagnostics

Gaps

  • Escher export: COBRA can export directly to Escher, the standard web-based metabolic network visualisation. RAVEN does not.
  • EFMviz: Visualising EFM composition, backbone networks, and enrichment.
  • Upset plots / epistasis heatmaps: Multi-model comparison and double-deletion visualisation.

14. Metabolomics integration

RAVEN

Limited to the production/uptake exchange bounds and ftINIT's optional metabolomics scoring block (which is not yet fully implemented in all pathways).

COBRA — metabotools

Function Description
prepIntegrationQuant Generates uptake/secretion profiles from extracellular metabolomics flux data
defineUptakeSecretionProfiles Calculates slope ratios for comparative conditions
setMediumConstraints Applies medium composition constraints
setQualitativeConstraints Applies detection-limit constraints from MS data
setSemiQuantConstraints Relative difference-based constraints between two conditions
setQuantConstraints Full quantitative metabolomics data integration as model constraints
setConstraintsOnBiomassReaction Applies growth rate constraints from doubling times
calculateLODs Converts detection limits (ng/mL → mM)
predictFluxSplits / computeFluxSplits Pathway contribution analysis for a metabolite
MetaboAnnotator Annotates metabolites via structure, InChI, VMH, PubChem, HMDB
conc2Rate Converts concentrations to uptake rates from cell parameters

Gaps

The metabotools suite is not present in RAVEN. For constraining models with extracellular metabolomics (uptake/secretion rates from NMR or LC-MS), COBRA provides a complete pipeline.


15. Model I/O and database connectivity

RAVEN

Area RAVEN
SBML Read/write SBML L2V1 and L3 (via importModel / exportModel)
Excel Structured Excel format for model editing (importExcelModel / exportToExcelFormat)
JSON COBRA-compatible JSON export
YAML RAVEN YAML format (human-readable, Git-friendly)
KEGG Full REST API integration: pathway, reaction, compound, enzyme queries
UniProt Gene/protein queries for annotation enrichment
HPA / GTEx Direct integration for tINIT expression data loading
Ensembl Gene ID mapping
Git-stable export exportForGit produces deterministic field ordering for clean diffs

COBRA

Area COBRA
SBML Read/write via libSBML (more complete SBML compliance including FBC v2)
Excel Read/write
JSON BIGG-compatible JSON
BiGG BiGG database REST API; metabolite/reaction annotation lookup
VMH Virtual Metabolic Human database integration
AGORA2 Direct interface to 7302 microbial GEMs
Recon3D Integration with the human metabolic reconstruction

Gaps

  • SBML compliance: COBRA's libSBML wrapper handles FBC v2 and Groups extensions more fully.
  • BiGG / VMH databases: COBRA is tightly coupled to these namespaces; RAVEN uses KEGG and does not have native BiGG/VMH lookup tools.
  • AGORA2 access: Not in RAVEN.

RAVEN strengths

  • YAML format and exportForGit: COBRA has no equivalent to RAVEN's Git-friendly export. The YAML format produces minimal, human-readable diffs — important for collaborative GEM development in version control.
  • Excel-to-model pipeline: RAVEN's Excel model format is mature and widely used in the community for manually curated models.
  • KEGG integration: RAVEN's KEGG functions are more comprehensive than COBRA's.

16. Summary gap table

Functions and capabilities present in COBRA but absent or minimal in RAVEN, grouped by priority for a GEM reconstruction lab.

High relevance for reconstruction labs

Capability COBRA functions RAVEN status
MILP gap-filling fastGapFill, swiftGapFill Only heuristic LP
Stoichiometric consistency fastcc, fastLeakTest, checkDuplicateRxn, detectDeadEnds Struct validation only
Multiple omics extraction algorithms GIMME, iMAT, FASTCORE, MBA, SWIFTCORE, eFlux ftINIT only
Multi-omics model extraction XomicsToModel Not present
Metabolomics constraints Full metabotools suite Not present
Double gene deletion doubleGeneDeletion Single KO only
pFBA (gene classification) pFBA Not present
Loopless FBA cycleFreeFlux, addLoopLawConstraints Not present

Medium relevance

Capability COBRA functions RAVEN status
Strain design (bilevel) OptKnock, optGene, GDLS Simplified approximation only
OptForce (must sets) findMustU/L/UU/UL/LL, optForce Not present
Genetic MCS calculateGeneMCS, calculateMCS Not present
MOMA / ROOM knockout prediction MOMA, ROOM, linearMOMA, linearROOM Not present
Markov-chain flux sampling ACHR, CHRR, RHMC Random-objective only
Thermodynamic constraints vonBertalanffy, thermoFBA Not present
EFM analysis EFMviz + EFMtool Not present
Conserved moieties conservedMoieties Not present

Specialised / niche

Capability COBRA functions RAVEN status
Microbiome community models mgPipe, SteadyCom Not present
Whole-body models Harvey/Harvetta toolbox Not present
Synthetic lethality Fast-SL Not present
Thermodynamic model setup setupThermoModel (vonBertalanffy) Not present
Escher / CellDesigner export Escher JSON, CellDesigner XML Not present

17. RAVEN strengths not in COBRA

These are areas where RAVEN has functionality that COBRA lacks:

Capability RAVEN functions COBRA status
KEGG-based de-novo reconstruction getKEGGModelForOrganism No equivalent pipeline
HMM-based gene calling getModelFromHomology (HMMER/BLAST) BLAST only via DEMETER
ftINIT (tissue-specific, HPA/GTEx) ftINIT, getINITModel2 Older methods only
Git-stable model export exportForGit Not present
YAML model format RAVEN YAML Not present
Flux path tracing traceFluxPath Not present
Flux diff between conditions compareFluxes Not present
Minimal medium (MILP) getMinimalMedium findMinimalMedia (similar)
Minimal flux (MILP, per-reaction) getMinNrFluxes sparseFBA (similar)
Reporter metabolites reporterMetabolites Not present
FSEOF FSEOF Not present
KEGG database queries getKEGGComps, getKEGGRxns, getRxnsFromKEGG, etc. Limited

Last updated: 2026-06-19. Based on RAVEN v2.x (develop3 branch) and COBRA Toolbox v3.x.