Testing cortical-state persistence as a dynamical mechanism of freezing of gait

Reproducible analysis code and frozen result summaries for the resubmission of manuscript rsif-2026-0121 to the Journal of the Royal Society Interface.

The repository analyses the public version-3 filtered multimodal freezing-of-gait dataset and separates three questions: concurrent EEG/FoG association, predictive value of continuous versus discretized representations, and prospective evidence for cortical-state persistence or antecedent spectral dynamics before FoG onset.

Data

Public source data:

Li et al., Multimodal Dataset of Freezing of Gait in Parkinson's Disease, Mendeley Data, version 3. DOI: 10.17632/r8gmbtv7w2.3.

The analysis expects the extracted Filtered Data directory from that release. The GitHub Actions workflow downloads the same public archive automatically.

Reproducibility

Python 3.12 is used in continuous integration. Install dependencies with:

python -m pip install -r requirements.txt

Run the complete reviewer-response pipeline with:

bash run_all.sh "/path/to/Filtered Data" work

The pipeline:

1. assembles and executes the recording-bounded core analysis;
2. executes the continuous prospective-onset analysis;
3. executes the strictly antecedent cortical-history analysis, in which the historical interval ends exactly where the current 2-s EEG window begins;
4. writes machine-readable result tables and reports below the requested output directory.

The release workflow also records the exact Python package versions used by GitHub Actions.

Frozen scientific results

The release is tied to the final resubmission analysis. The principal conclusions are:

• FoG-aligned spectral differences are heterogeneous across recordings under full-grid circular-shift nulls and movement sensitivities.
• EEG-only latent-state dwell does not show the hypothesized positive relation with FoG-onset risk; point estimates are non-positive across K=2--5 in the primary prospective risk set.
• Discretization into a three-state representation provides no robust predictive advantage over the underlying continuous feature representation.
• Strictly antecedent theta/low-beta activity shows temporal alignment with FoG onset under circular-shift nulls, especially for a 5-s preceding interval, but the joint participant-clustered effect is uncertain and does not improve participant-held-out prediction.

Compact machine-readable summaries used for the manuscript are stored in frozen_results/. Large per-window tables are generated by the workflow rather than committed to the repository.

Repository structure

• analysis/ -- analysis scripts used by the reproducibility workflow.
• frozen_results/ -- compact final result tables used to audit manuscript claims.
• .github/workflows/jrsi_review_reanalysis.yml -- clean-room download and rerun from the public dataset.
• run_all.sh -- local entry point.
• CITATION.cff -- software citation metadata.
• .zenodo.json -- metadata prepared for Zenodo archival.

Scope

This repository is a research reproducibility archive. It is not a clinical prediction system, diagnostic device, or treatment-control implementation.

Zenodo

This repository is prepared for a new versioned GitHub release and Zenodo archive corresponding to the resubmitted analysis. The previous Zenodo identifier associated with the earlier manuscript analysis is intentionally not used for this release. The new version DOI should be added here after Zenodo mints it from the GitHub release.