Novel Predictions: Results That Emerge Only from the Formal Model
The formal models developed in Chapters Energy Metabolism Models through Formal Causal Hierarchy Analysis produce fifteen predictions that range from formalizations of existing clinical intuition (e.g., damage prevention dominance) to genuinely counterintuitive claims (e.g., moderate patients benefiting most from mitochondrial supplements). Some of these predictions require the formal model to derive; others could be stated verbally but gain quantitative precision from the mathematics. None have been prospectively validated. Each is tagged with its biophysical origin equation and grouped by the type of insight it provides.
1 Predictions from the Multiplicative Structure of Metabolism
Metabolic subsystems are arranged in series (glycolysis → Krebs cycle → ETC → ATP synthase). The cascade example (Section Consolidated Cascade Example) demonstrates that two individually moderate deficits—35% Complex~I impairment and 30% NAD+ depletion—compound multiplicatively to produce 57% ATP reduction and 7-fold ROS elevation: $ 0.65 = 0.46$, not $ 0.65 + 0.70 = 1.35$ or \((0.35 + 0.30)\\/2 = 0.33\). This resolves the central clinical puzzle of ME/CFS: why patients are profoundly disabled despite no single biomarker being dramatically abnormal. The answer is that modest individual abnormalities multiply through the metabolic pipeline. Mono-omic studies measuring only one domain will systematically underestimate disease burden because they capture one factor in a product, not a sum.
The multiplicative structure has a direct treatment consequence: two interventions targeting different bottlenecks in the series pipeline (e.g., CoQ10 for ETC + nattokinase for oxygen delivery) produce super-additive effects (\(\mathcal{S} \approx 0.45\), Section Model-Predicted Treatment Candidates), while two interventions targeting the same bottleneck (e.g., CoQ10 + NMN, both acting on ETC/NAD+) show diminishing returns (\(\mathcal{S} \approx 0.12\)). This explains why supplement “stacking” sometimes produces dramatic results and sometimes produces nothing—it depends on whether the stack targets different or identical bottlenecks.
2 Predictions from Threshold Nonlinearities
The ATP synthase threshold equation (atp synthase) creates three distinct clinical populations. (1) Above the cliff (\(\alpha_\text{CI} > 0.80\), mild impairment): \(\Delta \Psi\) is well above threshold; mitochondrial supplements produce minimal benefit because ATP production is already near-maximal. (2) On the cliff (\(\alpha_\text{CI} \approx 0.60\)–$ 0.70$, moderate impairment): small improvements in Complex~I capacity produce disproportionate ATP gains because \(\Delta \Psi\) is near the threshold where the synthase driving force drops steeply. These patients respond best to mitochondrial support. (3) Below the cliff (\(\alpha_\text{CI} < 0.50\), severe impairment): \(\Delta \Psi\) is below threshold; the deficit is too large for supplements to bridge. Treatment requires either massive ETC restoration or bypass strategies (e.g., glycolytic support, ketone bodies). The counterintuitive prediction: moderate patients benefit most from CoQ10/NAD+ supplementation, not the most severely impaired.
The biogenesis equation (biogenesis) is a product of two Hill functions with opposing energy dependencies. AMPK activation (sensing energy deficit) says “build new mitochondria.” NAD+ depletion (also from energy deficit) blocks SIRT1-mediated PGC-1\(\alpha\) deacetylation, saying “cannot execute the build order.” The result is a non-monotonic hump: biogenesis peaks at intermediate energy deficit (where AMPK is active and NAD+ is still adequate) and collapses at severe deficit (where NAD+ depletion blocks execution). Below the threshold \(\gamma < 0.7\), the cell receives a maximal “build” signal but cannot act on it. This predicts that ME/CFS patients show elevated phospho-AMPK alongside reduced PGC-1\(\alpha\) activity—a decoupling signature. NAD+ precursor supplementation should improve biogenesis markers (mtDNA copy number, citrate synthase activity) before improving energy capacity, because it unblocks the rate-limiting step.
The GPR81 lactate feedback (Section Lactate Kinetics and Metabolic Flexibility) adds a second threshold: when \(\alpha_\text{CI} < 0.65\), the lactate → GPR81 → FFA suppression → more glycolysis loop gain exceeds unity, producing a bistable metabolic state. Patients below this threshold are locked into glycolytic dominance and cannot recover metabolic flexibility with mitochondrial support alone—they require either GPR81 antagonism or medium-chain triglycerides (which bypass CPT-I) in addition.
3 Predictions from Temporal Dynamics
The structural hysteresis (Equation hysteresis width) means that the separatrix between disease and health is asymmetric. To recover, \(\alpha_\text{CI}\) must be restored to a level above the value at which disease was triggered—not merely to the trigger value. Correcting biomarkers to pre-illness levels is necessary but not sufficient for recovery. The required overshoot grows with disease duration because epigenetic consolidation progressively widens the hysteresis gap. This explains the persistent clinical frustration: why doesn’t normalising the apparent biochemical deficits restore health? The model predicts that interventions producing temporary improvement followed by relapse are landing in the hysteresis gap—past the onset threshold but short of the recovery threshold.
The epigenetic consolidation dynamics (Equation intervention window) predict that early disease (\(< \tau_\text{window}\)) and late disease (\(> \tau_\text{window}\)) are qualitatively different problems. Early: the attractor is shallow, recovery requires only crossing the structural separatrix. Late: the attractor is deepened by methylation changes at disease-state CpG sites — whether gain or loss, both stabilize the disease attractor. The unified vector model (Chapter Formal Causal Hierarchy Analysis, Per-Locus Dynamics: Vector Model for Bidirectional Methylation) captures both directions: hypermethylation at ProA/gene-region loci deepens the attractor through stable gene silencing, while hypomethylation at ProB repeats deepens it through heterochromatin erosion and HSAT2 derepression (see Speculation Consolidation as Loss of Methylation, Chapter Causal Hierarchy: Root Causes, Amplifiers, and Consequences). Recovery requires both crossing the separatrix and reversing epigenetic changes on a months-to-years timescale. The window duration \(\tau_\text{window} \approx \ln(2) \\/ (k_\text{DNMT} \cdot overline(C)_\text{pro})\) is estimated at 3–12 months. This predicts a step-function-like relationship between disease duration at treatment initiation and treatment response—not a smooth gradient but an inflection where the problem changes character.
The piecewise recovery scaling (Table Disease Progression Models) predicts that the biological shadow—biomarkers improving weeks to months before any functional change is detectable—is an expected feature of the floor regime, not a treatment failure (Hypothesis The Biological Shadow: Biomarkers Improve Before Function). Damage prevention dominance (Hypothesis Damage Prevention Dominance Below the Severe Threshold) quantifies that a single prevented crash saves months to years of recovery time at severe and extremely severe levels.
4 Predictions from Oscillatory and Stochastic Dynamics
The critical slowing down analysis (Section Critical Slowing Down and Early Warning Signals) predicts specific statistical signatures in wearable data 24–48 hours before a PEM crash: rising lag-1 autocorrelation in resting heart rate, increasing HRV variance, and reduced HRV complexity. These are the generic early warning signals of an approaching bifurcation, applied to the energy envelope threshold. A smartphone app implementing rolling autocorrelation and variance detectors on HRV data could warn patients to reduce activity before the crash occurs. The same signals operating in reverse—paradoxical instability during a stable period—would indicate approaching recovery transition.
The Hopf bifurcation analysis (Section Endogenous Oscillations and Hopf Bifurcation) predicts endogenous symptom oscillations with a period of 2–6 weeks when the system resides near the bifurcation (Equation oscillation period). The oscillation period itself encodes disease dynamics: shorter period indicates higher immune–metabolic loop gain (more active disease); lengthening period signals approaching recovery. This transforms a subjective complaint (“my symptoms cycle”) into a quantitative biomarker.
The stochastic resonance analysis (Section Endogenous Oscillations and Hopf Bifurcation) predicts that for patients very near the separatrix, a non-zero level of physiological perturbation could in principle maximise spontaneous recovery probability. The optimal perturbation amplitude scales as \(\sigma^2_\text{optimal} \propto sqrt(\Delta U \\/ g_\text{loop})\). However, this prediction is in direct tension with the damage prevention dominance result (Hypothesis Damage Prevention Dominance Below the Severe Threshold): the ratchet model predicts that any perturbation carries a risk of irreversible ceiling loss \(\delta_k\), and the recovery cost of that loss is disproportionately high at lower \(B\). The risk-benefit asymmetry is stark—a failed perturbation (crash) causes permanent damage, while a successful one (separatrix crossing) is beneficial but uncertain. In practice, the stochastic resonance prediction applies only to the narrow subpopulation near the separatrix with shallow attractor depth (\(\Delta U\) small), where the perturbation needed is small enough that ratchet damage risk is negligible. For patients deep in the disease attractor or in the severe/extremely severe range, the damage prevention logic dominates and perturbation is contraindicated. This analysis does not endorse graded activity programs; the perturbations considered are endogenous physiological fluctuations (hormonal cycling, circadian variation), not externally imposed exercise.
5 Predictions from Network Structure
Network controllability analysis of the model Jacobian (Section Network Controllability and Minimum Intervention Sets) predicts that if the model topology is an adequate representation of the real biological network, the system requires a minimum of 4–6 independent driver nodes for full structural controllability. This is a property of the network topology, independent of parameter values, but dependent on whether the modelled connections are real. Furthermore, full controllability may not be the clinical goal—partial controllability (disrupting one critical feedback loop) might suffice for recovery. The minimum controllable driver sets span distinct subsystems: metabolic + immune + neuroendocrine + autonomic, at minimum. This explains why unselected monotherapy trials consistently show small average effect sizes even for drugs that may be highly effective within the right patient subtype.
6 Predictions from Comorbidity Coupling
The hEDS vascular coupling model (Section Extended Subsystem Couplings, Equation eds vascular) predicts that hypermobile Ehlers-Danlos syndrome imposes a permanent 10–20% reduction in the energy available for activity during upright hours, independent of any mitochondrial or immune dysfunction, purely from the increased sympathetic drive needed to maintain blood pressure against excessive venous pooling (\(\kappa = 1.5\) vs. 1.0 in normal connective tissue). This predicts that hEDS+ME/CFS patients should tend toward greater severity than non-hEDS patients with equivalent mitochondrial and immune profiles, that supine positioning should disproportionately benefit hEDS patients, and that compression garments should expand the energy envelope more in hEDS than in non-hEDS patients. The energy tax is posture-dependent and therefore modifiable—a directly actionable prediction.
7 The Repair Starvation Trap
At the bottom of the severity scale, Speculation Repair Starvation Trap: A Secondary Attractor at Very Low \(B\) identifies a potential secondary attractor where continuous chronic damage (ROS, inflammation) exceeds the diminished repair capacity even in the absence of discrete damaging events. Below a critical \(B_\text{trap}\), patients decline inexorably regardless of crash prevention. This predicts that a subpopulation of extremely severe patients deteriorates despite maximally protective environments—and that recovery from this regime requires interventions that bypass or dramatically enhance the endogenous repair machinery rather than merely preventing further damage.
None of the fifteen predictions listed here have been prospectively validated in ME/CFS cohorts. The multiplicative cascade, ATP cliff, and biogenesis paradox follow from established biochemistry applied to ME/CFS parameters, but the parameter values themselves are estimated from cross-sectional studies. The temporal predictions (intervention window, hysteresis, CSD) require longitudinal data that do not yet exist. The network controllability analysis depends on the model topology being an adequate representation of the real biological network. These are model-generated hypotheses, not established findings. Their value is that they are specific, falsifiable, and derived from biophysics rather than fitting—each can be tested with existing or near-future methods.