# C6 external-validation result and discordance diagnosis — 2026-10-30
## Conclusion: GSE263483 is not suitable as an external validation set for the GSE249311 signature
## (true biological difference, not a technical error)

## Validation results (GSE249311 vs GSE263483, both PyDESeq2 |LFC|>=0.58 & padj<0.05)
- A1:   rho=-0.057 (perm.P=0.001), doubly significant overlap 370, direction agreement 160/370 (43%)
- A2:   rho=-0.049 (perm.P=0.001), doubly significant overlap 180, direction agreement 69/180 (38%)
- T2DM: rho=-0.032 (perm.P=0.001), doubly significant overlap 418, direction agreement 191/418 (46%)

## Diagnosis (technical explanations excluded)
1. [diag1] Recomputed GSE263483 log2FC vs author-provided log2FC: Spearman rho=1.000 (n=31334)
   → the GSE263483 processing side is 100% trustworthy.
2. [diag2] The GSE249311 pipeline was independently validated by the C1-v2 anchor (39/39 mtDNA
   down-regulation, matching the preprint).
3. [diag3] 13 mtDNA genes: strongly down-regulated in GSE249311 (-1.4) but no such signal in
   GSE263483 (mean -0.01, 8/13 down) → a genuine biological difference between cohorts.

## Biological interpretation (for the Discussion)
- GSE263483 is a cohort specifically restricted to GDM-AGA (normal fetal growth); GSE249311 mixes
  outcomes and includes A1/A2/T2DM.
- The two cohorts differ in GDM treatment/severity/ancestry; genes up-regulated in GSE263483
  (SLC12A8/FSTL3/TMSB10, LFC 1.1-3.9, padj<1e-16) have no corresponding signal in GSE249311.
- This discordance is precisely the evidence basis for the "hyperglycemia effect vs macrosomia
  effect" decoupling design: an AGA-cohort signature ≠ a mixed-cohort signature.

## Disposition (Act)
1. GSE263483 is re-roleled as "decoupling arm 1" (hyperglycemia-effect reference) and no longer
   serves as external validation.
2. Validation strategy revised to: (i) GSE173193 pseudobulk (downloading) cross-dataset consistency;
   (ii) GSE70493 to be retried after the NCBI captcha cools down (the only independent cohort with
   literature-supported directional agreement with GSE249311); (iii) GSE267340 internal pseudobulk
   stability (leave-one-out).
3. GSE103552 was verified to be diabetic-pregnancy endothelial-cell data (dAEC/dVEC, no healthy
   controls) and removed from the validation list.
