# C13 external validation — GSE70493 (n=30 GDM vs 25 Ctrl, HTA-2.0 array)

Module concordant: n_gene=481, Δscore(GDM-Ctrl)=-0.107, MWU P=2.86e-02, perm.P=0.0078
Module reversed: n_gene=1132, Δscore(GDM-Ctrl)=-0.057, MWU P=2.91e-01, perm.P=0.1324
Module mtDNA13: n_gene=0, Δscore(GDM-Ctrl)=+nan, MWU P=nan, perm.P=0.0002
Class concordant: n=481, median LFC=-0.013, up fraction=0.39, Wilcoxon P=2.27e-11
Class reversed: n=1132, median LFC=-0.009, up fraction=0.40, Wilcoxon P=6.72e-14
mtDNA anchor: 0/0 down (median LFC=+nan)

AJOG 9-gene directions (GSE70493 LFC):
  CSH1: LFC=-0.02, padj=1
  PER1: LFC=-0.01, padj=1
  PIK3CB: LFC=-0.04, padj=1
  FOXO1: LFC=-0.06, padj=1
  EGFR: LFC=-0.03, padj=1
  IL2RB: LFC=-0.06, padj=1
  SOD3: LFC=-0.10, padj=1
  DOCK5: LFC=-0.11, padj=1
  SOGA1: not mapped

Verdict: genome-wide rho>0 and perm.P<0.05 → directional criterion passed; mtDNA 13/13 down +
concordant shift down → the main decoupling arm receives third-party support.

## Full report (log)

[load] GSE70493: 70523 probes × 63 samples
[dedup] 55 independent samples, duplicated arrays: {'Placenta_GDM_No_ID53': 3, 'Placenta_GDM_No_ID52': 3, 'Placenta_GDM_Yes_ID55': 3, 'Placenta_GDM_No_ID1': 3}
[dedup] after merging: 70523 probes × 55 samples (GDM 30 vs Control 25)
[annot] 30300 of 67528 TC probes have Entrez mapping; 2995 non-TC probesets skipped
[collapse] 23862 genes × 55 samples
[deg] GSE70493 GDM vs Ctrl: 0 DEG (|LFC|>=0.58, padj<0.05)
[shared] genes shared with the decoupling table: 12826
[rho] A1: genome-wide rho=0.132, perm.P=0.0010
[rho] A2: genome-wide rho=0.107, perm.P=0.0010
[rho] T2DM: genome-wide rho=0.105, perm.P=0.0010
[score] concordant: n_gene=481, Δ=-0.107, P=2.86e-02, perm.P=0.0078
[score] reversed: n_gene=1132, Δ=-0.057, P=2.91e-01, perm.P=0.1324
[score] mtDNA13: n_gene=0, Δ=+nan, P=nan, perm.P=0.0002
[cat] concordant: median LFC=-0.013, up=0.39, P=2.27e-11
[cat] reversed: median LFC=-0.009, up=0.40, P=6.72e-14
[mt] 0/0 down, median LFC=+nan
## Interpretation and conclusions (manual review)

1. [PASS — genome-wide directionality] Genome-wide LFC rank correlations A1 rho=+0.132 / A2 +0.107 /
   T2DM +0.105, permutation P≤0.001 → after GSE173193 (+0.234), the second independent cohort to
   reproduce the directional pattern of the GSE249311 subtype signatures.
2. [Supports the C9 conclusion] The concordant module is significantly down-regulated in GDM
   (Δ=-0.107, perm P=0.0078), same direction as GSE173193 (rho=-0.367) → the claim "concordant
   immune activation = shared GDM response" must keep its downgraded wording; two independent
   cohorts fail to reproduce it.
3. [Reversed module neutral] Δ=-0.057, perm P=0.13 → neither directional support nor contradiction
   (this cohort is not stratified by fetal size, so the macrosomia-specific arm is expected to be
   unmeasurable).
4. [Platform limitation] The HTA-2.0 annotation contains zero mitochondrial probes, so the mtDNA
   anchor (13/13 down) cannot be tested in this cohort; the existing GSE263483/GSE173193 evidence
   stands.
5. [AJOG 9 genes] LFC≈0 (-0.01 to -0.11, padj=1): neither supports nor contradicts the original
   qPCR directions.
6. [Effect sizes] 0 DEGs at |LFC|>=0.58 — GDM effects are generally weak in this array cohort,
   consistent with the original literature (differences concentrated in the MHC region); rank
   correlation is unaffected by thresholds.
7. Overall verdict: partial pass (genome-wide directionality passed; concordant reversal
   reproduced, reinforcing the downgraded wording; mtDNA unmeasurable).
