# C15 GSE173193 third-party validation v2 (correcting the GL-group misassignment)

## 1. Group-assignment correction

- GEO Series description: `2 placenta samples from PE, GDM, the elderly group and the control group
  respectively`
- **GL = advanced-age group**, not GDM, not LGA
- The original script used `GROUPS = {'C','G','P'}` + `GROUPS[smp[0]]` → GL1/GL2 start with "G"
  → merged into GDM
- Consequence: the original "GDM" group contained 4 samples (G1/G2/GL1/GL2), not 2; C9's
  rho=+0.234 and −0.367 were both built on the wrong grouping

  C1   -> Control       cells=7,829
  C2   -> Control       cells=4,800
  G1   -> GDM           cells=8,430
  G2   -> GDM           cells=6,161
  P1   -> PE            cells=11,187
  P2   -> PE            cells=5,192
  GL1  -> AdvancedAge   cells=7,568
  GL2  -> AdvancedAge   cells=7,410

- Sample count: 8 (the original script reported 6; actual 8)
- Group sizes: Control=2, GDM=2, PE=2, AdvancedAge=2
- CPM check: column sums=1e6 ✓ (the original script's sf≈1 never divided by library size — wrong units)

## 2. Directional concordance with the two decoupling arms (after group correction)

| Reference group | vs hyperglycemia arm rho (P) | vs macrosomia arm rho (P) |
|---|---|---|
| GDM (n=12,203 genes) | +0.024 (7.5e-03) | -0.272 (7.6e-206) |
| PE (n=12,203 genes) | -0.134 (4.9e-50) | +0.004 (6.6e-01) |
| AdvancedAge (n=12,203 genes) | -0.154 (4.7e-66) | -0.189 (6.5e-99) |

## 3. Reproduction of the signature sets in the two independent cohorts (core test)

| Signature | n genes | GSE173193-GDM rho vs hyperglycemia arm | GSE173193-PE rho vs hyperglycemia arm |
|---|---|---|---|
| reversed_by_macrosomia | 1,014 | +0.331 (P=2.5e-27) | -0.189 (P=1.2e-09) |
| concordant | 431 | -0.247 (P=2.1e-07) | -0.499 (P=1.8e-28) |
| macrosomia_assoc | 1,841 | -0.056 (P=1.7e-02) | -0.155 (P=2.5e-11) |

## 4. mtDNA anchor

- Mappable: 13/13
- GDM: mean -0.01, 8/13 down
- PE: mean -0.27, 13/13 down
- AdvancedAge: mean +0.47, 1/13 down

## 5. Comparison with the original version (quantifying the GL contamination)

| Metric | Original (GL merged into GDM, n=4) | Corrected (GDM n=2) | Change |
|---|---|---|---|
| reversed rho vs hyperglycemia arm | +0.230 | +0.331 | +0.101 |
| concordant rho vs hyperglycemia arm | -0.404 | -0.247 | +0.157 |

## 6. Verdict

- GL (advanced age) and GDM **cannot be treated as the same group** transcriptomically; after
  correction GDM has only 2 samples, so any GDM-related inference is descriptive only
  (n=2, no statistical power).
- If the reversed-set concordance had disappeared or sharply attenuated after correction, the
  original claim of "support for overgrowth-specific reversal" would have had to be deleted.

## 7. Manual interpretation (substantive changes after correction)

1. [Reversed is actually stronger] After correction, reversed (1,014) vs hyperglycemia arm rho
   rose from +0.230 to **+0.331** (P=2.5e-27). GL contamination **diluted** the signal rather than
   creating it. → The statement "the ECM-type reversed signature in an independent GDM cohort
   follows the hyperglycemia arm, not the overgrowth arm" **still holds after correction, and more
   robustly**.
2. [Concordant still fails] Corrected rho=-0.247 (P=2.1e-7), PE -0.499. Negative in both directions
   → "concordant = shared GDM immune activation" must be deleted or drastically downgraded.
3. [mtDNA is a PE feature] GDM -0.01 (8/13), **PE -0.27 (13/13 down)**, advanced age +0.47 (1/13).
   → Supports the methodological conclusion that "mtDNA down-regulation resembles a PE/mixed-cohort
   feature" (one of the three salvageable innovations of this project).
4. [PE and GDM are directionally opposite] PE vs hyperglycemia arm rho=-0.134, vs macrosomia arm
   +0.004; GDM shows the opposite pattern. → GDM and PE placental transcriptomes are systematically
   opposite in direction and must be clearly distinguished in the Discussion.
5. [Power] GDM n=2 after correction. All rhos are descriptive directional concordance only; the
   words "validation"/"replication" must not be used. Main-text wording: directional concordance
   (descriptive), not validation/replication.
