solver.press

Untested hypothesis

Feedback loops between physical tie reinforcement (\kappa) and local neuronal activity dynamics drive distinct hysteresis boundaries in neurodegenerative spreading processes, which can be mapped precisely using statistical dynamics inference.

NeuroscienceNov 12, 2026Evaluation Score: 70%

Adversarial Debate Score

61% survival rate under critique

Expert panel critique

Independent views, each critiquing the hypothesis on its own — the score rewards genuine disagreement and discounts consensus.

Gemini: Assessment: The hypothesis is highly falsifiable and theoretically robust, well-supported by network neurodynamics literature showing activity-dependent pathology spreading, multistability, and orthogonal tie-reinforcement parameters (\kappa). While the owner's validated experiments primari...
Mistral: The hypothesis is well-grounded in current literature on network spreading dynamics and neuronal feedback, with falsifiable claims and plausible mechanistic links. However, it lacks direct experimental validation from the owner’s work (which focuses on unrelated precision/optimization findings) a...
Claude: The hypothesis is conceptually coherent and partly grounded in the hysteresis/multistability and activity-dependent spreading papers, but it conflates two distinct frameworks: κ is the tie-reinforcement parameter from the temporal-network contact literature, not the structural coupling of protein...
ChatGPT: The excerpts support activity–spreading feedback and hysteresis, but do not establish that physical tie reinforcement κ drives distinct boundaries or that those boundaries can be inferred precisely; identifiability depends on observation conditions and competing mechanisms. The owner’s validated ...

Related patents (prior art)

This hypothesis overlaps subject matter covered by existing third-party patents. It is published as research, not as a patentable claim of ours.

Supporting Research Papers

Formal Verification

Z3 logical consistency:✅ Consistent

Z3 checks whether the hypothesis is internally consistent, not whether it is empirically true.

Experimental Validation Package

This discovery has a Claude-generated validation package with a full experimental design.

Precise Hypothesis

In a computational model coupling activity-dependent synaptic/structural reinforcement (parameterized by coupling strength κ, representing activity-driven strengthening of physical ties such as axonal transport pathways or trans-synaptic connectivity) to a pathological spreading process (e.g., prion-like tau/α-synuclein propagation), there exists a critical range of κ (κ_c1 < κ < κ_c2) within which the system exhibits bistability: two distinct stable states (low-burden "healthy" and high-burden "degenerated") coexist for the same external parameter values (e.g., seed pathology load, baseline activity level), producing a hysteresis loop with measurable width ΔΘ > 0 in the control parameter space. This hysteresis width and its boundaries (κ_c1, κ_c2) can be estimated from time-series data (neuronal activity + connectivity/atrophy proxies) using statistical dynamics inference (e.g., maximum-likelihood fitting of stochastic differential equations or Ising-like kinetic models) with parameter recovery error <15% on held-out simulated data, and the inferred hysteresis boundaries will differ measurably (effect size Cohen's d > 0.5) between simulated "high-feedback" (κ large) and "low-feedback" (κ small) regimes.

Disproof criteria:
  • No bistable regime (single-valued steady-state burden as function of control parameter) found across the entire physiologically plausible κ range (tested via continuation/bifurcation analysis over κ ∈ [0, κ_max]).
  • Inferred hysteresis width ΔΘ statistically indistinguishable from zero (95% CI includes zero) across ≥3 independent model parameterizations and ≥500 simulated trajectories.
  • Statistical dynamics inference fails to recover ground-truth κ and hysteresis boundaries in simulated data (parameter recovery error >40%, or inference likelihood surface is flat/non-identifiable).
  • In empirical (animal/human) data, no corresponding signature (e.g., path-dependence of pathology burden on prior activity/connectivity history) is detectable above noise floor after controlling for confounds (age, comorbidity, scan-to-scan variability).
  • Effect size between high- and low-feedback regimes Cohen's d < 0.2 (negligible) in properly powered comparison (n≥30 simulated networks per condition).

Spine & Adversarial ReadReady for validation

“This hypothesis is testing whether activity-dependent structural reinforcement feedback (κ) produces a statistically detectable, inferable hysteresis boundary in network-based neurodegenerative spreading models and in longitudinal human pathology data.”

  • highExisting static-network diffusion models (e.g., Raj's network diffusion model, Iturria-Medina's epidemic spreading model) already fit tau/amyloid spreading patterns reasonably well without invoking activity-dependent feedback or hysteresis — why is the added complexity of κ-feedback and bistability necessary rather than overfitting?
    The EVP partially addresses this via model comparison (ΔBIC, cross-validated predictive performance) against static-network null models, but this comparison is not explicitly specified as a required step in METHODOLOGY — this is a gap that should be added as an explicit model-selection step (static diffusion vs. feedback model) before claiming the feedback mechanism is necessary.
  • highClinical PET/MRI sampling intervals (6-12 months) are far coarser than the timescales (ms-hours) over which activity-structure feedback is hypothesized to operate; detecting a hysteresis signature from such sparse, indirect, noisy burden measurements may be fundamentally underpowered regardless of inference method sophistication.
    Acknowledged explicitly in KNOWN_FAILURE_MODES and tested via Abort Checkpoint 4; however, no power analysis is yet provided to quantify minimum detectable effect size given ADNI's actual sampling density — this should be computed via simulation (sampling the synthetic model at ADNI's real visit schedule) before committing to Phase 2 spending, and is currently a methodological gap.
  • mediumWhy statistical dynamics inference (particle MCMC/SDE fitting) specifically, rather than simpler alternatives like direct bistability detection via clustering/mixture models on cross-sectional data, or machine-learning classifiers trained directly on trajectories? The choice of methodology is not justified against simpler baselines.
    Not resolved in current EVP — a baseline comparison (simple mixture-model/dip-test approach vs. full dynamical inference) should be added as a required ablation in METHODOLOGY to justify the added complexity and computational cost (1800 GPU-hours) of the full SDE inference pipeline versus cheaper alternatives that might achieve comparable detection power.

Experimental Protocol

Phase 0 (in silico, 4-6 weeks): Build minimal agent-based/mean-field model with N=500-10,000 nodes on empirically-derived connectome topology (e.g., Human Connectome Project structural matrices), activity dynamics (Wilson-Cowan or leaky integrate-and-fire), structural reinforcement rule dκ_ij/dt = f(activity_i, activity_j, κ_ij), and spreading/damage process (SIR-like or diffusion-reaction with activity-dependent rate). Sweep κ (global scaling) and seed pathology location/magnitude; perform forward+backward parameter continuation to detect hysteresis. Apply statistical dynamics inference (e.g., Bayesian filtering, SDE parameter estimation via EM or variational inference) to recover κ and boundaries from simulated noisy observations (activity timeseries + coarse-grained "burden" signal at sparse time points mimicking PET/MRI sampling).

Phase 1 (ex vivo/animal validation, 3-6 months): Use organotypic brain slice cultures or mouse models (e.g., tau-seeding models, APP/PS1) with chronic calcium imaging + post-hoc connectivity/pathology staining at multiple timepoints; manipulate activity (chemogenetics/optogenetics) to test for path-dependence predicted by model.

Phase 2 (human retrospective, 6-12 months): Apply inference framework to existing longitudinal ADNI-type datasets (tau-PET, amyloid-PET, structural MRI, optionally resting-state fMRI) at ≥3 visits per subject, test for signature of hysteresis (bimodal burden distribution at matched baseline risk, path-dependence on early activity/connectivity measures).

Required datasets:
  • Human Connectome Project (HCP) structural connectivity matrices (diffusion MRI tractography-derived) for realistic network topology.
  • Simulated ground-truth datasets generated internally (synthetic activity + structural + pathology time series, N≥1000 networks x multiple κ regimes).
  • ADNI (Alzheimer's Disease Neuroimaging Initiative): longitudinal tau-PET (AV-1451), amyloid-PET, structural MRI, and where available resting-state fMRI, ≥3 visits, n≥300 subjects with full multimodal coverage.
  • Animal model data: tau-propagation mouse model (e.g., from Jucker lab-style seeding paradigms) with chronic in vivo calcium imaging — may require new data collection or collaboration.
  • Organotypic slice culture electrophysiology/imaging data (new data collection, ~6 months).
  • Software/models: Virtual Brain (TVB) platform or custom Wilson-Cowan network simulator; PyMC/Stan or custom SDE inference toolkit (e.g., based on Kalman/particle filtering).
Success:
  • Bifurcation analysis confirms bistable region exists for realistic parameter ranges (κ within literature-supported bounds) in ≥80% of tested network topologies.
  • Statistical inference recovers ground-truth κ and hysteresis boundary location with mean absolute error <15% and 90% CI coverage ≥85% across ≥500 synthetic trials.
  • Effect size (Cohen's d) between high- and low-feedback regime hysteresis widths >0.5, p<0.01 (bootstrapped, n≥30 per group).
  • In ADNI data, evidence of bimodal or path-dependent burden distribution consistent with hysteresis (Hartigan's dip test p<0.05, or mixture model outperforms unimodal by ΔBIC>10) in ≥2 independent cohort subsets.
  • Model predictions of individual "distance to tipping point" correlate with subsequent 2-year clinical progression (AUC >0.65) in retrospective validation.
Failure:
  • Bifurcation analysis shows monotonic, single-valued steady-state response across entire tested κ range — no bistability.
  • Inference pipeline parameter recovery error >40% or CI coverage <60%, indicating non-identifiability.
  • No statistically significant difference (d<0.2, p>0.2) between high/low feedback regime hysteresis widths.
  • ADNI data shows no bimodality/path-dependence signature (dip test p>0.3) and mixture model does not outperform unimodal null.
  • Predicted tipping-point proximity does not correlate with clinical progression (AUC ~0.5).

1,800

GPU hours

270d

Time to result

$45,000

Min cost

$320,000

Full cost

ROI Projection

Commercial:

Creates IP-able diagnostic/prognostic algorithm (software-as-medical-device candidate) licensable to neuroimaging/diagnostics companies (e.g., companies developing tau-PET analytics); generalizable methodology (statistical dynamics inference for hysteresis detection) applicable beyond AD to Parkinson's, ALS, and potentially non-neurological network-mediated diseases (epidemic spread on social/power networks), broadening platform value; attractive for partnership with pharma running anti-tau/anti-amyloid trials seeking better patient stratification to improve trial power (estimated trial cost savings of 15-30% via better enrichment).

🔓 If proven, this unlocks

Proving this hypothesis is a prerequisite for the following downstream discoveries and applications:

  • 1clinical-tipping-point-biomarker-panel
  • 2personalized-intervention-timing-model-for-AD
  • 3generalized-hysteresis-framework-for-other-network-diseases

Prerequisites

These must be validated before this hypothesis can be confirmed:

  • prion-like-spreading-validated-in-connectome-models
  • activity-dependent-synaptic-plasticity-rate-constants-calibrated

Implementation Sketch

# Phase 0: Model + bifurcation
define node state: (activity x_i, local burden p_i)
define edge state: kappa_ij
dx_i/dt = -x_i + sigmoid(sum_j kappa_ij * x_j) - gamma*p_i
dkappa_ij/dt = eta*(x_i*x_j - kappa_ij)   # Hebbian-like reinforcement
dp_i/dt = alpha * sum_j kappa_ij * f(p_j - p_i) * g(x_i)  # activity-gated spreading

for kappa_global in sweep_range(0, kappa_max, step):
    simulate_forward(seed_load=low -> high)
    simulate_backward(seed_load=high -> low)
    record steady_state_burden(kappa_global)
plot_hysteresis_loop(); detect_bistable_region(kappa_c1, kappa_c2)

# Inference pipeline
def infer_params(observed_trajectories, prior):
    model = StateSpaceModel(transition=ODE_system, observation=noisy_sparse_sampling)
    posterior = particle_MCMC(model, observed_trajectories, prior, n_particles=2000)
    return posterior.kappa, posterior.hysteresis_bounds, credible_intervals

# Validation
for trial in synthetic_trials:
    true_params = sample_prior()
    obs = simulate_and_sample(true_params, noise_model, sampling_schedule=ADNI_like)
    est_params = infer_params(obs)
    record_error(true_params, est_params)

# Human data application
for subject in ADNI_cohort:
    traj = extract_longitudinal_burden(subject)  # tau-PET SUVR over visits
    est = infer_params(traj, prior=population_prior)
    distance_to_tipping = compute_distance(est.kappa, est.hysteresis_bounds)
    correlate(distance_to_tipping, clinical_outcome_2yr)
Abort checkpoints:
  • Checkpoint 1 (Week 4): If bifurcation analysis shows no bistable region across realistic parameter space in base model — abort/redesign model before further investment.
  • Checkpoint 2 (Week 10): If inference pipeline fails parameter recovery on synthetic data (error >40%) after two rounds of pipeline refinement — abort or pivot to simpler inference target (e.g., just κ estimation without full boundary mapping).
  • Checkpoint 3 (Month 4): If connectome-scale model (HCP topology) fails to reproduce bistability seen in minimal model — reassess topology assumptions before proceeding to empirical data phase.
  • Checkpoint 4 (Month 7): If ADNI retrospective analysis shows no dip-test significance and no mixture-model improvement after full preregistered analysis — abort clinical translation claims, report null result.

Source

AegisMind Research
Need AI to work rigorously on your problems? AegisMind uses the same multi-model engine for personal and professional use. Get started