Every number here comes from the program's code and results, not a summary slide — and every one is reproducible. The headline: enforcing one law of physics exposes large, systematic, cross-organism errors in the way cells are modelled today.
For each glycolytic reaction, the equilibrium constant implied by the model's kinetics versus the one demanded by thermodynamics. Where they cross 1.0 in opposite directions, the model is running the reaction backwards.
| Enzyme | ΔG° (kJ/mol) | K_eq · kinetic | K_eq · thermo | Discrepancy | Direction |
|---|---|---|---|---|---|
| Aldolase FBA | +22.4 | 5.12 | 1.0e-4 | 42,886× | ↔ reversed |
| Lactate dehydrogenase LDH_L | -26.8 | 0.970 | 48,743 | 50,217× | ↔ reversed |
| Phosphoglycerate mutase PGM | +4.5 | 31.00 | 0.160 | 192× | ↔ reversed |
| Triose-phosphate isomerase TPI | +5.6 | 11.34 | 0.110 | 107× | ↔ reversed |
| G3P dehydrogenase GAPD | +4.6 | 6.01 | 0.160 | 38× | ↔ reversed |
| Phosphoglycerate kinase PGK | -22.1 | 64.70 | 2421.00 | 37× | ✓ agrees |
| Phosphofructokinase PFK | -14.6 | 322.90 | 361.10 | 1.1× | ✓ agrees |
| Enolase ENO | +1.2 | 2.65 | 4.65 | 1.8× | ✓ agrees |
K_eq · kinetic is implied by the published model's forward/reverse rates; K_eq · thermo is computed from eQuilibrator standard Gibbs energies. Source: the program's detailed-balance reversibility audit.
Lactate dehydrogenase is where the disagreement is sharpest — and where it is testable. The two models predict opposite metabolic fates for the same cell.
The test — Measure lactate and pyruvate in JCVI-syn3A culture supernatant by LC-MS or NMR. A ratio ≫ 1 confirms the thermodynamic model; a ratio near 1 supports the kinetic one.
The biology — Mycoplasma are textbook lactate fermenters — their primary carbon waste is lactate, exactly what the thermodynamics predicts and the existing model does not.
Run the gold-standard audit on three independent, separately-built genome-scale models and the wrong-direction rate lands in a tight 13–17% band — and clears a shuffled-K_eq null (~40%) by many standard deviations every time.
| Organism | Domain | Reversed (D4) | Central metab. | z-score | p-value |
|---|---|---|---|---|---|
| JCVI-syn3A 455 genes · 338 rxns | Prokaryote · minimal cell | 16.8% 21/125 | 27.3% | -7.91 | 2.5×10⁻¹⁵ |
| E. coli iML1515 1,515 genes · 2,712 rxns | Prokaryote · fully-capable | 16% 16/100 | 40% | -4.66 | 3.1×10⁻⁶ |
| S. cerevisiae · Yeast-GEM v9 ~6,000 genes · 4,131 rxns | Eukaryote · 14 compartments | 13% 13/100 | — | -5.79 | 6.9×10⁻⁹ |
7,181 reactions audited at the structural tier; 325 at the D4 gold standard shown above. Negative z indicates far fewer thermodynamically-consistent reactions than chance. Source: cross-organism universality analysis.
The whole thesis rests on which behaviours fall out of the physics versus which we put in by hand. So we publish the distinction, line by line — including what isn't wired in yet.
The central axiom of the framework — by construction, not emergent.
Appears with the invariant ON; disappears under a detailed-balance-violated null control.
No death logic is written; collapse follows from the chemistry — reproduce it live in the simulator.
Division at the Young-Laplace instability from measured constants — but timescale calibration is uncertain; reported with two options side-by-side.
Independent datasets (syn3A, E. coli, yeast); significant against a shuffled-K_eq null.
Partly recapitulates the proteomics used to initialise it; quantified with a uniform-IC control.
Ribosome pool is held static in this version; a dynamic module is built but not yet wired in.
Implemented separately; not yet coupled to the main cell-cycle loop.
* Emergent, but the timescale calibration is uncertain; reported with two calibration options side-by-side.
That is the bridge between a strong computational result and a validated one — and where backing makes the difference.