# C17 figure set v2 (fixing figure–text number inconsistencies, review item P0-2)

## Corrections
1. Fig2: removed the plotting-stage second filter (baseMean>10); DEG counts now come directly from
   the complete DEG tables
   → A1 3,950 (1,368/2,582), A2 1,306 (666/640), T2DM 3,717 (1,586/2,131),
   fully consistent with the text (the v1 figures showed 3,103/1,147/3,175, contradicting the text).
2. Fig2 PCA: unified to log2(CPM+1) + top 2,000 variable genes → PC1=30.8%, PC2=14.8%
   (the v1 figure's 47.6% came from log1p(raw counts) all genes; the text's old value 24.4% came
   from log2(counts+1) top-2000 — the three specifications disagreed; now unified, text uses 30.8%)
3. Fig2 caption states the specification: "log2(CPM+1), top 2,000 variable genes"
4. Fig3: switched to the corrected 03b table (true CPM, Y/CNV excluded,
   macrosomia_only → macrosomia_assoc)
   → y-axis range shrinks from ±32 to ±10 (extreme LFCs disappear after artifact removal)
5. Fig5A: switched to 06b per-sample Fisher-z pooling (previously 16 samples pooled across spots)
   → S3-Endothelial changes from (old) +0.37 to +0.10; S3-CTB +0.42, S2-Fibroblast +0.40
6. Fig1: all numbers updated to corrected values; explicitly labeled "descriptive, not
   identifiable"; "sex inference: cases F, control M"; MR labeled "not Bonferroni-significant"
7. Fig4B title adds "sex-confounded", Fig4C adds "descriptive; n=4"

## Outputs
figures/Fig1_study_design.png … Fig5_spatial_MR.png (300 dpi, all regenerated)
scripts/09b_publication_figures_v2.py (original 09 retained as audit comparison)
