Campaign run · Parkinson's disease

parkinson-asyn-propagation-r2

PDCompleted the pipelineAttempt 2 of 2 at this topic

Does a Susceptible-Infected-Removed (SIR) compartmental model of α-synuclein spread between brain regions, parameterised by Fullard 2024 multi-region snRNA-seq cell-type fractions and SNCA expression, plus Allen Human Brain Connectivity Atlas projection edges, reproduce the canonical Braak staging order (DMNX → SNpc → cingulate → M1 → V1) at qualitatively correct relative timing?

Not supported 4
Inconclusive 1

Open the final report (PDF)This run in the hypothesis explorer

Status

How far it got: Completed the pipeline
Tested at least one hypothesis and reached the multi-agent review stage. This shows the run went through the whole pipeline, not that its work passed a quality bar.
Hypotheses: 5 nominated; 5 evaluated.
Recorded status: Halted. The run recorded a final status of halted: it was closed without being marked done. A halted run can still have tested hypotheses and written a final report.
Census class: Success. Census class is a label from an earlier, process-level census of the campaign. It does not map one-to-one onto the stages above. How the classes compare.

Research question

The sentence under the run id is quoted from this brief.

Full research brief given to the run

Braak staging (Braak 2003 Neurobiol Aging) is the most-cited spatial-temporal claim in PD neuropathology, but its mechanistic explanation — that α-syn spreads cell-to-cell along a connectome gradient, modulated by selective vulnerability — has been tested in mouse atlases (Henderson 2019, Zheng 2019) but not at human cell-type granularity. Fullard 2024 provides the missing per-region cell-fraction and SNCA-expression priors.

Hypotheses (pre-registered):

H1 (primary — Braak rank-order recovery): An SIR + connectivity model with SNCA-expression-weighted per-region susceptibility recovers the rank-order of predicted regional peak-prevalence times consistent with Braak rank, with Spearman ρ ≥ 0.7 across the 5 modelled regions (DMNX, SNpc, cingulate, M1, V1). Bootstrap 95% CI of ρ excludes 0.

H2 (secondary — SNCA-weighting necessity): Ablating the SNCA-expression weighting (uniform susceptibility across regions) degrades rank-order recovery by ≥ 0.3 in Spearman ρ. (Test that cell-specific susceptibility is necessary, not sufficient.)

H3 (secondary — cingulate cell-type prediction): The model predicts which cingulate cell types reach infection threshold first; we pre-register "excitatory layer-5/6 neurons" as the top prediction. The top-3 model-predicted cell types overlap ≥ 2 with the top-3 Lewy-burdened cell types reported by Goralski 2024 LAMDA (Zenodo 10.5281/zenodo.10729767).

H4 (secondary — vagal-route ablation): Removing the DMNX → SN projection edge (parameter ablation) shifts predicted peak-time rank-order by > 1 rank position. A falsifiable mechanistic claim about the vagal spread route.

H5 (secondary — connectome-only vs. cell-autonomous): Compare three models — (A) connectome-only SIR, no cell-type weighting; (B) cell-type-weighted SIR (this proposal's primary); (C) cell-autonomous-only (each region's I trajectory independent of others). Report which best fits Braak rank-order. A finding that (A) or (C) wins is an informative inversion of the spread-only hypothesis.

Null payoff. If no SIR variant recovers Braak rank-order, the connectome-only hypothesis is incomplete and pure cell-autonomous selective vulnerability (independent of spread) dominates. Informative as a head-to-head test of two competing field-level hypotheses.

The research question comes from the topic's current research brief. All attempts at a topic share it; earlier attempts may have started from an earlier wording.

Hypotheses and verdicts

Each hypothesis was written with its decision rule before any analysis (or, for on-the-fly hypotheses, recorded during execution). AI adjudication agents applied that rule to the run's latest outputs and quoted the evidence verbatim. Where the data did not meet the rule, the verdict follows the data, even if the run's own label or report claimed support.

Each hypothesis as pre-registered (a condensed statement with its decision rule), the adjudicated verdict, the verbatim evidence and the adjudicator's note.
ID	Hypothesis and decision rule	Verdict
H1Pre-registered	SNCA-weighted SIR (model B) recovers proxy Braak rank order [1,3,5,5,6] of t_peak across 5 regions (rho>=0.7, bootstrap CI lower bound >0).	Not supportedIn the explorer
Evidence (verbatim from the run outputs) not meeting the pre-registered criterion that requires CI exclusion of zero Adjudication note Latest EXECUTE stage 19: rho=0.711 (meets point threshold), CI [-0.304, 1.000], perm p=0.094; n=5 regions, plan conceded ~50-65% power -> borderline INCONCLUSIVE.
H2Pre-registered	Removing SNCA weighting (model A, connectome-only) degrades rho by >=0.3 relative to model B.	Not supportedIn the explorer
Evidence (verbatim from the run outputs) the observed Δρ is negative and the CI does not exclude zero. H2 is NOT SUPPORTED. Adjudication note Delta rho(B-A) = -0.207 (CI -1.304 to 0.913); connectome-only A outperforms B (0.918; 1.000 when independently optimised, post hoc).
H3Pre-registered	Within PFC (cingulate proxy) excitatory L5/6 neurons reach infection threshold first; model top-3 overlaps >=2 with LAMDA Lewy-burdened top-3 (Fisher p<0.05).	InconclusiveIn the explorer
Evidence (verbatim from the run outputs) H3 is not biologically interpretable; the following analysis is reported for pre-registration completeness only. Adjudication note OmiCairn headline 'H3 is NOT SUPPORTED' (overlap 0, Fisher p=1.0), but report itself deems it uninterpretable (DMNX_Neu/GPI_Neu annotation artefacts in PFC; sub-SIR never reached threshold).
H4Pre-registered	Ablating DMNX->GPi edge shifts peak-time rank by >1 position for >=1 region.	Not supportedIn the explorer
Evidence (verbatim from the run outputs) Zero rank displacement (max \|Δrank\| = 0) was observed across all five regions after removing the DMNX→GPi edge (Fig. 5a). Adjudication note None recorded.
H5Pre-registered	Model-family comparison: connectome-only (A), SNCA-weighted (B, primary), cell-autonomous (C) ranked by Braak rho; A or C beating B reported as-is.	Not supportedIn the explorer
Evidence (verbatim from the run outputs) The three model variants ranked as: Model A (ρ = 0.918, CI [0.745, 1.000]) > Model B (ρ = 0.711, CI [−0.304, 1.000]) > Model C (ρ undefined; Fig. 5b,c,d). Adjudication note UNCERTAIN mapping: H5 is descriptive ('always reportable'); verdict reflects that the primary model B did not rank best (inversion). Pairwise Delta CIs include 0; alternative INCONCLUSIVE.

Final report

Title (as written by the system)

Connectome topology (rho = 1.00) outperforms SNCA expression for Braak alpha-synuclein staging: a pre-registered human atlas benchmark

Main conclusion

The pre-registered SNCA-expression-weighted SIR model failed H1 (rho = 0.711, 95% CI -0.304 to 1.000, permutation p = 0.094, n = 5 regions), and H2, H3 and H4 were also not supported. Connectome-only models did better (rho = 0.918 at shared parameters; rho = 1.000 when optimised post hoc), and Dijkstra graph distance alone gave rho = 0.975 (p = 0.015). The report reads this as the Allen connectivity topology encoding the Braak gradient, a confound for atlas-based propagation analyses.

Report titles are the system's own and can state a result more strongly than the report's verdicts support. Read the main conclusion and the verdicts with it. The main conclusion was written during release review and keeps the report's own numbers and verdicts.

The PDF is published exactly as released. It may call itself a “paper”, call the system's internal multi-agent review “peer review”, and cite internal stage folders by number.

Open the final report (PDF, 347 KB)

All 2 attempts at this topic

parkinson-asyn-propagation-r1Tested, stopped before review
parkinson-asyn-propagation-r2Completed the pipeline Report

Attempts are numbered in launch order. A topic was relaunched when an earlier attempt stopped or was superseded.