# C14 decoupling analysis v2 (artifact/normalization/filter/sex-confounding corrections)

## 1. Input and normalization corrections

- Pseudobulk: 4 samples × 36591 gene symbols (raw counts aggregated)
- Library sizes: GDMM_GDMM1=115,407,305, GDMM_GDMM2=103,554,273, GDMMINS_GDMMINS=67,791,303,
  CONTROL_CONTROL=104,488,418
- CPM check: column sums = 1.000e6 ✓ (the original script's column sums reached 2.5e12 — wrong units)
- Original table CPM_mean_Control max = 2.501e+12 → after correction = 25,438

## 2. Per-sample fetal-sex inference (Y-chromosome genes vs XIST)

                 Y_marker_counts  XIST_counts inferred_sex
sample
GDMM_GDMM1                    24        14129       female
GDMM_GDMM2                    12        18191       female
GDMMINS_GDMMINS            26300         8209         male
CONTROL_CONTROL            29124        10731         male

**Key conclusion**: GDMM1/GDMM2 (Y≈0, XIST high) = female fetuses; CONTROL (high Y expression) =
male fetus. Cases and control of the overgrowth arm are **completely separated by fetal sex**;
any GDMM-vs-Control difference is indistinguishable from a sex effect.
This is a **fundamental design flaw** of this arm that cannot be fixed by statistical correction;
it can only be labeled non-identifiable.

## 3. Expression filter (two-arm CPM>=1 and raw count>=10)

- Genes before filtering: 36,591
- Genes after filtering: 15,575 (42.6% retained)
- Reference: the original script's max(CPM)>=20 under wrong units ≈ ≥1 count, i.e. effectively no filter

## 4. Artifact-gene exclusion (Y chromosome + CNV/segmental-duplication loci)

- Y-chromosome genes (gene_info chr=Y): 88
- Common CNV loci (curated): 20
- Segmental-duplication gene families (regex): 76
- Total slated for exclusion: 184
- Of these, passing the expression filter: 29

## 5. Hyperglycemia arm (GSE263483, n=8v8, valid DESeq2 inference)

- DEG table: 31,310 genes
- Significant (|LFC|>=0.58 & padj<0.05): 3,590

## 6. Category counts under the three specifications

| Specification | Usable genes | hyp_sig | concordant | reversed | macrosomia_assoc |
|---|---|---|---|---|---|
| (a) Original 03 (wrong CPM, artifacts included) | 13,443 | 719 | 500 | 1,172 | 2,930 |
| (b) Corrected CPM + pseudocount + filter | 13,940 | 848 | 513 | 1,164 | 2,214 |
| (c) Fully corrected (+ artifact exclusion) | 12,401 | 749 | 439 | 1,027 | 1,872 |

## 7. Artifact composition of the original macrosomia_only (n=2,930)

- Y-chromosome genes: 8
- CNV/segmental duplication: 11
- Total artifact genes: 19 (19/2,930 = 0.6%)
- Artifacts among the original |LFC_mac| top-20: 12/20

## 8. Threshold sensitivity (fully corrected specification)

| |LFC| threshold | concordant | reversed | macrosomia_assoc |
|---|---|---|---|---|
| 0.29 | 667 | 1,397 | 4,031 |
| 0.58 | 439 | 1,027 | 1,872 |
| 1.0 | 156 | 558 | 854 |

## 9. Leave-one-out (each case vs the same control)

| Case | n genes (filtered) | concordant | reversed | macrosomia_assoc | Jaccard vs pooled (reversed) |
|---|---|---|---|---|---|
| GDMM_GDMM1 | 12,401 | 362 | 1,376 | 3,557 | 0.703 |
| GDMM_GDMM2 | 12,401 | 788 | 665 | 1,458 | 0.554 |

(If the two cases' reversed sets disagree strongly, the class is single-case-driven rather than a
reproducible programme.)

## 10. Genome-wide direction-agreement rate (no longer claimed as "decoupling")

- Direction agreement = 45.0% (5,582/12,401); binomial test vs H0=50% P = 1.14e-28
- But because the macrosomia arm has n=1 control and complete sex confounding, this deviation
  cannot be attributed to biology: sex confounding alone can generate systematic directional shifts
  (Y genes excluded, but X dosage, cell composition and individual effects remain).
- Conclusion: **the direction-agreement rate cannot be used as evidence of "decoupling"**; reported
  descriptively only.

## 11. Position of the mtDNA anchor in the two arms

- Mappable mtDNA genes: 13/13
- Hyperglycemia arm mean LFC = -0.01 (8/13 down)
- Macrosomia arm mean LFC = +0.29 (3/13 down)
- ⚠ mtRNA is sensitive to RNA integrity/rRNA depletion and belongs to technically sensitive genes,
  see P1-4.

## 12. Enrichment analysis (the original manuscript's claims had no results file; computed here)

### macrosomia_assoc (n=1872)
  - Regulation Of Cell Migration (GO:0030334): adj.P=3.69e-07, overlap=84/434
  - Positive Regulation Of Cell Migration (GO:0030335): adj.P=4.60e-07, overlap=60/272
  - Positive Regulation Of Cell Motility (GO:2000147): adj.P=4.60e-07, overlap=52/221
  - Transmembrane Receptor Protein Tyrosine Kinase Signaling Pathway (GO:0007169): adj.P=5.43e-07,
    overlap=61/284
  - Positive Regulation Of Protein Phosphorylation (GO:0001934): adj.P=6.00e-06, overlap=71/377
### reversed_by_macrosomia (n=1027)
  - Extracellular Matrix Organization (GO:0030198): adj.P=2.69e-09, overlap=36/176
  - Collagen Fibril Organization (GO:0030199): adj.P=1.31e-08, overlap=17/42
  - Supramolecular Fiber Organization (GO:0097435): adj.P=5.89e-07, overlap=45/316
  - Skeletal System Development (GO:0001501): adj.P=3.27e-05, overlap=26/149
  - Extracellular Structure Organization (GO:0043062): adj.P=8.71e-05, overlap=21/109
### concordant (n=439)
  - Defense Response To Virus (GO:0051607): adj.P=1.62e-05, overlap=20/189
  - Defense Response To Symbiont (GO:0140546): adj.P=3.25e-05, overlap=17/148
  - Negative Regulation Of Viral Process (GO:0048525): adj.P=5.96e-05, overlap=11/61
  - Regulation Of Viral Genome Replication (GO:0045069): adj.P=1.08e-03, overlap=10/67
  - Positive Regulation Of Cytokine Production (GO:0001819): adj.P=1.19e-03, overlap=22/320

## 13. v2 conclusions (for rewriting the main text)

1. [P0-3 fixed] CPM units corrected, column sums=1e6; pseudocount changed to +1 CPM; filter changed
   to two-arm CPM>=1 & count>=10.
2. [P0-1 fixed] 19 of the 2,930 macrosomia-set genes (0.6%) are Y-chromosome/CNV/segmental-
   duplication artifacts; after full correction the class shrinks to 1,872 genes.
3. [P0-5 exposed] All cases are female fetuses and the control is male — complete sex confounding
   → the overgrowth arm is **non-identifiable**; no difference can be attributed to fetal
   overgrowth.
4. [P0-6 fixed] concordant/reversed gene counts are highly sensitive to thresholds and to
   single-case drives (see §8/§9).
5. [P0-7 basis updated] concordant reverses in both GSE173193 and GSE70493 → cannot be called a
   "shared immune-activation programme".

## 14. Supplementary diagnosis: chromosome distribution (X-dosage assessment)
- Genome-wide X-chromosome baseline: 3.3%
- macrosomia_assoc: X=63/1,872 (3.4%), enrichment 1.02x, Fisher P=0.89 → **no X enrichment**
  → the set is not a simple X-dosage effect, but sex confounding can still act through cell
  composition/individual effects and cannot be excluded.
- reversed_by_macrosomia: X=51/1,027 (5.0%), enrichment 1.50x, Fisher P=0.0034 → mild X enrichment
  (recomputed 2026-11-01: baseline = remaining 11,374 analyzed genes, 410/12,401 = 3.31%; two-sided Fisher test)
- concordant: X=18/439 (4.1%), enrichment 1.24x, P=0.34 → none
- Composition of the original macrosomia_only |LFC| top-20: 7 Y genes
  (DDX3Y/EIF1AY/KDM5D/RPS4Y1/UTY/USP9Y/ZFY) + GSTM1 (CNV) +
  NOTCH2NLB/TBC1D3D/AGAP6/FAM99A/FAM163B/TMEM88B/LRRC55 (segmental duplications/low expression)
  + CGB5/CGB8 (placenta-specific multicopy family) + DEFA3 (CNV) + SOX17
  → **12 of the top-20 are clear artifacts**; the remainder are low-expression or high-copy
  family genes.

## 15. Enrichment results (first actual computation here, replacing unsupported statements)
- macrosomia_assoc: Regulation Of Cell Migration (adj.P=3.7e-7), Positive Regulation Of Cell
  Motility (4.6e-7), RTK signaling (5.4e-7) → **"cell migration" reproduces under the original
  specification**
- reversed_by_macrosomia: ECM Organization (2.7e-9), Collagen Fibril Organization (1.3e-8)
  → consistent with the "ECM/collagen" narrative, but note the fibroblast-composition confounding
  (P0-10)
- concordant: Defense Response To Virus (1.6e-5), Response To Type II Interferon (2.9e-3),
  T Cell Receptor Signaling (2.9e-3) → in fact an **antiviral/IFN response**, not generic
  "immune activation"
- The original "DNA replication / axon guidance" did **not** enter the top-8 in this computation
  and should be deleted.
