Abstract Genome-scale metabolic models have progressed from stoichiometric reconstructions to predictive, constraint-aware platforms. In this review, we organize strategies for multi-omics integration not by data type but by the constraint logic they impose on model solution spaces. Biomass functions enforce composition and maintenance demands, while transcriptomic switches prune network feasibility. Enzyme and expression valves cap flux capacity, proteome budgeting enforces allocation trade-offs, and thermodynamics and fluxomics provide physical and experimental calibration. Machine learning bridges infer missing priors while retaining mechanistic structure. These categories translate into practical workflows, spanning enzyme-constrained modeling, thermodynamic embedding, and fluxomics-guided calibration, together with minimal reporting standards to ensure transparency and reproducibility. Emerging directions include the integration of single-cell and spatial data, physics-informed and graph-based machine learning, and translational pipelines that couple computational predictions with experimental validation. By framing omics integration through constraint architectures, this review provides a coherent agenda for making GEMs reproducible, portable, and biologically meaningful across biotechnology, medicine, agriculture, and environmental applications.
Shahreen et al. (Mon,) studied this question.