Curating MitoCore: A Standardized Small-Scale Human Metabolic Model as Platform for Proteomics Integration and Disease Modeling
Emanuel LangeiD, Alejandro Benjamin Reyes SantamariaiD, Robert HeyeriD
Abstract
MotivationCentral human metabolism powers cellular processes, yet its dysregulation in disease remains poorly understood. While comprehensive genome-scale metabolic models like Human-GEM are available, their size limits interpretability and computational efficiency. Conversely, the smaller MitoCore model is more manageable but lacks the standardized annotations and curated gene-protein-reaction (GPR) associations necessary for omics integration like protein-constrained modeling. Improving MitoCores annotation quality is therefore essential for its use in integrative workflows. ResultsWe systematically updated MitoCore to enhance compatibility with the protein-constrained modeling framework sMO-MENT. By restructuring legacy annotations and integrating data from Human-GEM and MitoMammal, we increased EC-codes from 354 to 593 and UniProt-annotated genes from 0 to 592. MitoCore captures central metabolic processes, confirmed by mapping its reactions to 51 of 106 metabolic KEGG modules. Integration of thrombocyte proteomics and experimental ATP data for original and curated models showed an increase in mapped proteins (228 to 294) and reactions with kcat values (295 to 310), adding 33 protein-constrained reactions. Consequently, prediction errors for exchange fluxes and ATP production decreased by 19% and 89%, respectively, with 100% of ATP predictions falling within the 95% confidence interval (compared to 16% for the original model). Finally, we implemented a continuous integration/continuous deployment pipeline for automated updates from future Human-GEM releases. These improvements provide a computationally efficient, well-annotated model for studying central metabolism across human cell types. Availability and ImplementationAll source code for reproducing results from this paper is available at https://doi.org/10.5281/zenodo.20813825.
02 /Specification & Classification
Click any tag to browse matching papers- Citations
- 0
- Organism
- Assay
- Research Focus
- Approach
In-depth summary
Claude Opus 5.5 · xhigh effort · 8 Nov 2026 · from the full text of version 2 (30 Jun 2026) · CC BY 4.0
Re-annotated MitoCore (SBML3, UniProt, EC, Human-GEM gene rules) supports sMOMENT proteomics integration, reducing thrombocyte ATP prediction error by 89% relative to the original model.
Genome-scale Human-GEM (12,931 reactions) is hard to inspect manually and costly for analyses such as flux sampling. The compact MitoCore (555 reactions) models proton motive force explicitly but lacked the standardized annotations that automated omics integration needs, so this work adds them, along with gene rules kept in sync with Human-GEM, to enable protein-constrained modeling.
Key results
ShownA four-stage curation (SBML3 conversion, UniProt retrieval, MitoMammal updates, Human-GEM transfer) greatly expanded MitoCore's machine-readable annotations and improved its overall quality score.
EC codes rose from 354 to 593 and UniProt-annotated genes from 0 to 592 (Original vs MitoCore Human-GEM). Step 4 alone added 208 genes, 236 EC codes and 545 isoenzymes. The MEMOTE total score rose from 35% to 65%. · Abstract; G. The Curation Workflow enriches MitoCore with…
ShownCuration increased the proteomics and kinetic data that autoPACMEN could attach, which gave more enzyme-constrained reactions. The uncurated Original model could not be processed at all because it lacked UniProt accessions.
Mapped proteins: 228 (of 387 with UniProt accession) for Preliminary and MitoMammal vs 294 (of 597) for Human-GEM. Reactions with at least one kcat: 295, 298, 310. Protein-constrained reactions: 168, 170, 201 (Preliminary, MitoMammal, Human-GEM). 33 protein-constrained reactions were added overall. · Abstract; I. Standardized Annotations facilitate the…
SuggestedProtein-constrained, curated models reproduced Seahorse-measured glycolytic and OXPHOS ATP production in resting thrombocytes far better than the unconstrained Original model, including the glycolytic index.
ATP RMSRE was 0.085 to 0.095 (protein-constrained) vs 0.819 (Original). Fraction of samples inside the measured 95% CI (MCP): 0.996 (MitoMammal), 1.000 (Human-GEM), 0.771 (Preliminary), 0.163 (Original). Glycolytic index was 0.237 to 0.240 vs 0.241 measured. · Abstract; I. Standardized Annotations facilitate the…; Tables
ShownEnzyme constraints shrank the feasible flux space and modestly improved exchange-flux predictions. Annotations derived from Human-GEM added no benefit beyond the MitoMammal updates, and no model captured the measured acetate uptake.
Protein-constrained models kept about 66% of the Original solution space (mean flux range 38 vs 58 mmol/gdwh). Exchange RMSRE: 0.631 (MitoMammal), 0.632 (Human-GEM), 0.645 (Preliminary), 0.800 (Original). Only the Original model predicted notable acetate uptake, an order of magnitude below the measured value. · I. Standardized Annotations facilitate the…; Discussion; Tables
ShownMapping reactions to KEGG modules shows that MitoCore covers core energy metabolism but only a subset of human metabolic modules. A few mitochondrial modules are covered better than in Human-GEM.
MitoCore covered 51 and Human-GEM 105 of 106 human KEGG modules (Human-GEM subset of 9,634 cytosolic, mitochondrial and extracellular reactions). Human-GEM contains 100% of MitoCore genes, 72.7% of its reactions and 94.6% of its metabolites. TCA cycle, mitochondrial fatty acid elongation and glycine cleavage modules are covered better in MitoCore. · Abstract; H. KEGG Module Mapping supports MitoCore’s…; Discussion
Methods, data and reproducibility
MitoCore was curated in four stages (SBML3 annotation parsing, UniProt IDs, MitoMammal updates, Human-GEM annotations/GPRs). Stages were parameterized for resting thrombocytes, protein-constrained with sMOMENT (autoPACMEN) where annotations allowed, calibrated to Seahorse data, and compared by FVA and hit-and-run flux sampling. A weekly GitLab CI/CD job reruns stage 4.
- Original MitoCore model of human central and mitochondrial metabolism (SBML2), the starting point for curation · human · n = 555 reactions, 441 metabolites, 391 genes
- Human-GEM genome-scale model, used as the source of annotations and GPR rules · human · n = 12,931 reactions, 8,461 metabolites, 2,848 genes
- MitoMammal model updates: DHODH-linked CoQ reduction reactions, corrected complex I and IV GPRs, UCP1 gene · mammal (MitoMammal maps human to murine genes) · n = 5 new reactions
- Public quantitative platelet proteomics (copies/cell) from isolated donor thrombocytes, converted to mmol/gdw using 2.13 pg dry weight · human
- Public Seahorse dataset of thrombocyte ATP production (glycolysis and OXPHOS), used for calibration (mean 1.707 mmol/gdwh) · human
- Public 13C metabolic flux data for resting platelets, giving glucose, acetate and lactate exchange bounds · human
- Existing human thrombocyte metabolic model iAT-PLT-636, used to set the exchange boundary conditions · human
- Human KEGG metabolic modules retrieved via API for coverage mapping · human · n = 106 modules
- Methods & tools
- COBRApy (SBML2 to SBML3 conversion), autoPACMEN toolbox (sMOMENT protein constraints, kcat calibrator), MEMOTE test suite, Flux balance analysis and flux variability analysis, PolyRound, hopsy (uniform coordinate hit-and-run sampling), BRENDA (version 2024_1), SABIO-RK, BiGG (version 1.6), MetaNetX (version 4.5), KEGG, HGNC and UniProt APIs, GitLab CI/CD pipeline with GitLab Runner
- Data & accessions
- https://github.com/SysBioChalmers/Human-GEM
- Software & environment
- SBML3 version 1 via COBRApy. Database versions: BiGG 1.6, BRENDA 2024_1, MetaNetX 4.5. Human-GEM 2.0.0 is described as the current release. Python package versions and containers are not specified.
- Assay conditions & compute
- Most computations ran on a Windows 10 laptop (Intel i5-1145G7, 16 GB RAM). Flux sampling (40,000 samples, 16 Markov chains) ran on a Linux server (AMD EPYC 7543 32-Core, 512 GB RAM). Runtimes are not reported.
- Reproducibility gaps
- The calibration-candidate algorithm, modified calibrator and error metrics appear only in the supplementary information. The ATP-ase kcat sensitivity result is given as data not shown. The authors note that autoPACMEN depends on proprietary MATLAB licenses.
Limitations (6)
- The proteomics, 13C and Seahorse inputs come from different donors and media (Tyrode's with glucose and acetate vs DMEM with glucose and glutamine), which may explain the missed acetate uptake. The 13C data also lacked donor metadata, so exchange-flux errors have no 95% confidence intervals. · stated by the authors
- Errors in Human-GEM GPRs propagate into the curated model. For example, the SUCOASm rule merges the genes for the ADP-forming and GDP-forming subunits, so MitoCore Human-GEM may be unsuitable where high mapping specificity is needed. · stated by the authors
- Constraints use only total protein abundance, so post-translational regulation such as PDH phosphorylation is not captured. Swapping proteomes across conditions assumes kcat values stay roughly constant. · stated by the authors
- Models were calibrated to a single resting condition, so whether they generalize to activated or dysregulated metabolism is unclear. Seahorse ATP estimates also rely on simplifying assumptions about media and cell handling. · stated by the authors
- Contrary to the authors' expectation, Human-GEM enrichment did not further lower prediction errors, and its solution space was nearly equal in size to the other protein-constrained models. They suggest a shared rate-limiting ATP-ase kcat may constrain total ATP production in all models. · stated by the authors
- ATP predictions are scored against the same Seahorse data used for kcat calibration and for the sampling lower bound, so ATP accuracy is not an independent validation. Only one cell type was tested. · noted in this summary
Who should read:Read this if you need a compact, well-annotated human central metabolism model for proteomics-constrained flux analysis, or an example of keeping a small model automatically in sync with Human-GEM releases.
Reproducibility ledger
Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.
Sign in to file a reproducibility report.
Related papers
- MitoCore: A Curated Constraint-Based Model For Simulating Human Central Metabolism
Anthony C. Smith, Filmon Eyassu, … Alan J. Robinson·BMC Systems BiologyPeer-reviewed·Nov 24, 2017
- MitoMouse is a model reconstruction of murine mitochondrial metabolism
Stephen P. Chapman, Thibaut Molinié, … Bianca Hermine Habermann·bioRxiv·Jan 2, 2023
- MitoMAMMAL: a genome scale model of mammalian mitochondria predicts cardiac and BAT metabolism
Stephen P. Chapman, Theo Brunet, … Bianca Hermine Habermann·Bioinformatics AdvancesPeer-reviewed·Nov 3, 2024
- Network reconstruction and modelling made reproducible with moped
Nima P. Saadat, Marvin van Aalst, Oliver Ebenhöh·MetabolitesPeer-reviewed·Mar 22, 2022
- A pipeline for the reconstruction and evaluation of context-specific human metabolic models at a large-scale
Vítor Vieira, Jorge Ferreira, Miguel Rocha·PLoS Computational BiologyPeer-reviewed·Jun 24, 2022
- Integrative Modeling and Analysis of Fungal Central Carbon Metabolism
Janaka N. Edirisinghe, Claudia Lerma‐Ortiz, … Christopher S. Henry·bioRxiv·May 12, 2026