Scripts for Supplementary Table S5 of the Parkinson disease companion paper (McCaulley, locked 468-gene panel, PD)
Rebuilt 14 October 2026 (agreement-check tolerance corrected 15 October 2026) from the Table S5 legend; they reproduce every value in Table S5.

Files
  pd_s5.py        reproduction of the locked PD meta-analysis and sensitivity analyses A-D (Table S5)
  limma_py.py     Python implementation of the GEO2R two-group limma model (lmFit/eBayes, robust = FALSE)
  pd_s5_expected_output.txt   output of pd_s5.py

Requirements: Python 3 with pandas, numpy, scipy and openpyxl.
Inputs (place in raw/ beside the scripts; all public):
  GEO series matrices: GSE49036, GSE20141, GSE7621, GSE8397-GPL96 (A chip), GSE8397-GPL97 (B chip)
  GEO platform tables: GPL570-55999.txt, GPL96-57554.txt, GPL97-17394.txt
  MASTER_panel_FINAL_v468_7.5.26.xlsx (Zenodo 10.5281/zenodo.22178094)
  PD_per_gene_meta_analysis_468panel_v1.2.csv (Zenodo version 1.2; locked values, for comparison)
Run: python3 pd_s5.py

Contrasts (locked): GSE49036 PD Braak 3-6 (15) vs Braak 0 control (8); GSE20141 PD 10 vs control 8; GSE7621 PD 16 vs
control 9; GSE8397 all A-chip arrays (superior frontal gyrus, lateral and medial substantia nigra), 29 PD vs 16 control
arrays (15 and 7 donors), disease control MS155 excluded. Probe sets annotated to more than one gene symbol are excluded.
Locked rule: limma fitted on all probe sets of each array; lowest-P probe set per panel gene; SE = |log2FC / t|;
DerSimonian-Laird random effects; BH FDR across meta-analysed panel genes; high-confidence = FDR < 0.05, k >= 3,
direction concordance >= 70%, nominal P < 0.05 in >= 2 cohorts (cohort P = 2*Phi(-|t|), or the limma moderated-t P).
Sensitivity analyses (cohort P from the moderated t):
  A  per-sample mean of all probe sets of the gene; limma fitted on the panel genes' gene-level values
  B  the probe set with the highest average expression (chosen without reference to the PD-control comparison),
     taken from the all-probe-set fit
  C  A with GSE8397 restricted to substantia nigra, averaged per donor (15 PD, 7 control donors), HG-U133B probe sets added
  D  B with the same GSE8397 changes
Also: GSE8397 donor-level SN only; A or B with donor-level SN; locked + B chip; and the contrasts as described in the
earlier version of the paper (GSE49036 Braak 5-6 vs 0; GSE8397 lateral SN, 9 vs 7, A chip).

Verification (14 October 2026): all 12 summary rows of Table S5 reproduce exactly (genes, FDR < 0.05, high-confidence,
retained, additional); the locked 56-gene high-confidence set is identical to Zenodo v1.2; 418 of 419 locked genes are
recovered (AFDN is annotated as MLLT4 in GPL570-55999); per-gene log2FC and FDR in the Table S5 Per-gene sheet agree to
rounding, and high-confidence status in A-D and the n-of-4 robustness count agree for all 63 genes.
