Standardization of human genome-scale metabolic models for fluxomics and systems biology applications
Loading...
Date
2026-10-21
Authors
Παντζίρη, Μαριλένα
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Human genome-scale metabolic models (GEMs) provide systematic representations of human metabolism and support flux analysis and multi-omics integration. Yet human GEMs have evolved through successive generations of genome annotation, biochemical databases, identifier systems and reconstruction practices. Because many reconstructions originated when these resources and their interconnections were less mature, individually valid annotations may coexist within the same model reaction without describing mutually consistent biological relationships. Existing standards and quality-control frameworks improve structure and annotation coverage, but do not systematically test whether enzyme, protein, reaction and metabolite annotations converge on the same biochemical transformation.
The main objective of this thesis was therefore to make cross-layer annotation consistency testable against a common representation of the human metabolic reactome. Four specific aims were pursued: (1) to construct an enzyme-centric reference connecting enzyme activities, reviewed human proteins, biochemical transformations, metabolites and stoichiometry, while retaining directionality and subcellular localization as biological context; (2) to characterize the reconstructed reactome as a quantitative knowledge landscape; (3) to evaluate cross-layer consistency using the Human2 model (https://github.com/SysBioChalmers/Human-GEM) as a proof of concept and (4) to derive general annotation requirements and curation guidelines for the reproducible comparison of human GEMs.
A reference human metabolic reactome was reconstructed by integrating curated resources, principally Rhea, UniProt, ExplorEnz, ChEBI, KEGG and MetaCyc, with Reactome and other resources retained as supporting cross-reference and pathway evidence. Biochemical transformations were defined by chemical participants and stoichiometric relationships rather than identifiers alone; directionality and localization were represented separately as biological-context properties, and source provenance was preserved. The reference comprises several thousand retained EC classifications, associated reviewed human proteins, a source-derived reaction set, participating metabolites and a separately generated transport layer. Its characterization showed frequent many-to-many relationships between enzyme activities and reviewed human proteins, and an uneven distribution of evidence across the reactome. A cross-layer-complete core supported by all evidence layers coexists with regions in which the underlying knowledge remains incomplete or conflicting.
Human2 was used as a highly curated proof of concept. The analysis did not assess biological correctness or predictive performance; it tested whether annotations attached to the same
reaction were mutually consistent with the reference. Cross-layer consistency was evaluated at two levels: the biochemical transformation, at which the enzyme classification, the protein, the reaction and the metabolite participants and their stoichiometry were examined, and the biological context, at which directionality and subcellular localization were examined against the reference. Identifier correspondence was not treated as sufficient evidence of equivalence. Candidate mappings were evaluated through independent biological and chemical evidence layers, and the outcomes were retained separately, so that support, incompleteness, ambiguity, conflict and cases outside the scope of the reference remained distinguishable. The application showed that recognized identifiers and successful model-level quality control do not necessarily guarantee cross-layer consistency. Of the transformations that remained unresolved, 73.7% still carried at least one principal annotation type. Therefore, the unresolved set is not an unannotated region of the model, and the distinction between insufficient or non-convergent evidence and missing annotation cannot be made by counting identifiers alone. The resulting guidelines extend identifier- and model-level standardization by requiring consistency among chemical identity, stoichiometry, reaction-enzyme relationships and enzyme-protein support, while treating directionality and localization as separate biological-context layers and preserving missing, ambiguous or conflicting evidence as distinct states. The principal contributions are therefore a reference, provenance-preserving human metabolic reactome and a general framework for testing annotation consistency across biological layers. Future work should apply the framework to additional human GEM lineages, maintain the reactome as a versioned resource and formalize the guidelines into explicit annotation consistency criteria.
Description
Keywords
Human genome-scale metabolic models (GEMs), Metabolic model standardization, Human reference metabolic reactome, GEM annotation, Cross-layer consistency in GEM annotation, Systems biology, Fluxomics