Primer: genome-scale metabolic models (GEMs)¶
Background reading for anyone coming to this project from nutrition rather than
systems biology. biology-as-code is not a GEM and does not run flux balance
analysis — but GEMs are the established modelling tradition this work sits next
to, and the vocabulary shows up constantly in the surrounding literature.
What a GEM is¶
A GEM (genome-scale metabolic model, also called a genome-scale metabolic reconstruction or network) is a mathematical model of an organism's entire known metabolism, built from its genome.
It contains:
- All (or nearly all) known metabolic reactions the organism can perform
- The genes and proteins encoding the enzymes for those reactions, via GPR rules (gene–protein–reaction associations)
- Metabolites and their stoichiometry
- Compartments (cytosol, mitochondria, extracellular space, and so on)
The model is represented as a stoichiometric matrix — the S-matrix — and analysed with constraint-based methods, most commonly flux balance analysis (FBA).
When people say "run a GEM" or "constrain the GEM with a diet," they mean: take this genome-scale network of reactions, set bounds on the exchange reactions based on the food a person ate plus any other physiological constraints, then compute which fluxes through the network are possible.
The main models you'll see referenced¶
| Model | What it is | Reference |
|---|---|---|
| Recon3D | The main human GEM | Brunk et al., Nat Biotechnol 2018 — PMID 29457794 |
| AGORA | GEMs for 773 gut bacteria | Magnúsdóttir et al., Nat Biotechnol 2017 — PMID 27893703 |
| AGORA2 | Expanded to 7,302 microorganisms | Heinken et al., Nat Biotechnol 2023 — PMID 36658342 |
| Harvey / Harvetta | Whole-body models (WBM) joining organ-level human GEMs via blood compartments, male and female | Thiele et al., Mol Syst Biol 2020 — PMID 32463598 |
| VMH | The Virtual Metabolic Human database hosting these GEMs plus metabolite, reaction, gene, and food data | Noronha et al., Nucleic Acids Res 2019 — PMID 30371894 |
Whole-body models are the reason this vocabulary matters to nutrition: they integrate metabolism, physiology, and the gut microbiome into a single personalisable object, which is the closest existing analogue to what a mechanistic "what did this meal do" model needs.
How this project relates¶
biology-as-code works at a different altitude. A GEM answers which fluxes are
feasible across the whole network; this package models what happens to a meal
through named, inspectable stages — digestion machines, gates and bounds, and
teaching pathway graphs — with provenance attached to every value.
The two are complementary rather than competing. The GEM tradition supplies the mechanistic ceiling; the contribution here is the provenance discipline described in FDP-1 and the constitution, so that a number's origin and evidence grade travel with it instead of being lost on the way into a model.
A 2005 source map for the reconstruction-database generation (EcoCyc, MetaCyc, KEGG, and which printed URLs are dead) is in david-nielsen-2005.md.