# C9 third-party directional validation (05, GSE173193) — 2026-10-30
## Data
- GSE173193 placenta scRNA (10x): C1/C2 control, G1/G2 GDM, P1/P2 PE → pseudobulk per sample
- Median-of-ratios size factors → CPM → log2 LFC (descriptive, n=2 per group)

## Results
- Genome-wide: GDM-LFC vs hyperglycemia arm rho=-0.105, vs macrosomia arm rho=-0.210
  (cross-platform noise dominates, consistent with C6)
- **Key supporting evidence**: within the reversed gene set (n=1155), GSE173193-GDM correlates with
  the hyperglycemia arm at rho=+0.234 (p=8e-16)
  → an independent GDM cohort (no macrosomia selection) follows the "hyperglycemia arm" direction
  rather than the "macrosomia arm", i.e. the macrosomia-arm reversal is a macrosomia-specific
  effect, not platform noise
- Concordant gene set (n=492): GSE173193-GDM vs hyperglycemia arm rho=-0.367 → the immune-activation
  axis of the shared GDM response does not hold in this independent cohort; the claim must be
  downgraded or explained as cohort difference in the paper
- mtDNA anchor: GDM +0.27 (no down-regulation, consistent with AGA/macrosomia arms), PE -0.36
  (PE shows a down-regulation tendency)
  → the mtDNA suppression signature fits PE/mixed-cohort characteristics better, further supporting
  "not an obligatory GDM effect"

## Limitations
- Fetal-growth outcomes of the GSE173193 GDM samples unknown (LGA status unclear); macrosomia
  confounding cannot be fully excluded
- n=2 per group, no statistical inference; scRNA depth is low and LFC magnitudes are compressed
