Can we trust genome-scale metabolic models?

Genome-scale metabolic models (GEMs) have become one of the most mature computational frameworks in systems biology and are increasingly being integrated into multi-scale and whole-cell modeling approaches. While the reliability of metabolic network reconstruction has improved substantially through advances in automated curation, thermodynamic modeling and enzyme-constrained formulations, accurately predicting intracellular metabolic fluxes remains a major unresolved challenge. Unlike network topology, flux distributions depend on multiple interacting assumptions, including environmental constraints, objective functions, contextualization strategies and the intrinsic degeneracy of constraint-based models. This review examines whether current genome-scale metabolic models can be considered reliable predictors of intracellular metabolic states. Recent developments in context-specific reconstruction, enzyme-constrained modeling and flux sampling are discussed, highlighting both their methodological advances and their remaining limitations. Particular attention is devoted to the often-overlooked interplay between intracellular enzymatic constraints and extracellular nutrient availability, which may substantially influence the effective information content of enzyme-constrained models in human systems. The persistent lack of comprehensive benchmark datasets is also examined, emphasizing that most current validation strategies assess biological hypotheses rather than predicted flux distributions themselves. Overall, the evidence suggests that genome-scale metabolic models are currently more robust as mechanistically informed frameworks for interpreting multi-omics data than as direct predictors of quantitative intracellular fluxes, while also identifying the key methodological challenges that must be addressed before genome-wide flux prediction can become routinely reliable.