parkinson-multimodal-ppmi-trajectory-r1

Using only the publicly summarised longitudinal endpoints from PPMI — CSF α-syn-SAA Vmax/Tmax/AUC group means (Russo 2025 Lancet Neurology supplement), Olink Explore top-50-protein medians (Bayati 2024 Acta Neuropathologica supplement), motor MDS-UPDRS- III group-mean curves (Marek 2018 + downstream PPMI publications), free-water dMRI per-visit means (Burciu 2023 Frontiers Neurology supplement), LRRK2-G2019S SAA-stratified curves (Simuni 2024 Movement Disorders supplement) — can a Gaussian-process latent-variable model (GPLVM) infer a 1-D disease-time axis whose ordering recapitulates clinically known PD progression milestones?

PPMI raw individual-level data are under a Data Use Agreement (DUA). Trajectory inference papers that re-derive from raw data cannot be reproduced without the DUA. This proposal asks the inverse methodological question: how much disease-trajectory reconstruction is possible from supplementary-table summary statistics alone? If the answer is "a lot", this is a reproducibility win and a template for PPMI-DUA-free research. If "very little", it puts a quantitative upper bound on summary-table-only PD work and justifies the DUA application.

Hypotheses (pre-registered):

H1 (primary — disease-time correlates with time-from-diagnosis): The GPLVM-inferred 1-D latent disease-time axis correlates with published per-visit time-from-diagnosis at the cohort-summary level with Spearman ρ > 0.6 (n ≥ 50 visit-aggregated points across baseline / Y1 / Y2 / Y3 / Y5 timepoints × multiple cohort-subgroup strata).

H2 (secondary — modality leave-one-out generalisation): Free-water dMRI (Burciu 2023) is held out; GPLVM trained on SAA + Olink + MDS-UPDRS only recovers free-water dMRI rank-order with ρ ≥ 0.4. (Tests that disease-time is modality-general, not modality-specific.)

H3 (secondary — LRRK2 leftward-shift): LRRK2-G2019S carriers (Simuni 2024) have a leftward-shifted inferred disease-time distribution vs. sporadic PD at matched clinical-stage (Cohen's d > 0.5, two-sample Welch t-test p < 0.05). Consistent with the SAA-negative LRRK2 subtype being molecularly distinct.

H4 (secondary — modality-specific weight check): Report per-modality GPLVM loading magnitude on the primary latent factor. No single modality should contribute > 70% of the loading. If one does, the "multi-modal" claim is misleading and that modality alone would do.

H5 (secondary — supplementary-table fetchability audit): Of the 6 cited supplements (Russo, Bayati, Hällqvist, Burciu, Simuni, Marek), how many are (a) machine-readable (TSV / XLSX), (b) PDF-table only, (c) image only / not fetchable? This number itself is an informative methodological finding about PD trajectory work reproducibility.

Null payoff. If summary tables don't encode enough information to reconstruct disease time (H1 ρ < 0.4), the PPMI raw-data DUA application is justified — and the proposal would put a hard number on what summary-table-only PD trajectory work can deliver.

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.