# C10 MR design and QC (updated) — 2026-10-30
## Data-source change log (Check)
- Public eQTLGen text files contain only Z scores, no Meta-Beta/Se (verified on the 2019-12-11 FDR
  release) → Wald/IVW not directly possible
- The SMR_formatted package is binary BESD (besd/epi/esi) with no official Windows parser → abandoned
- Final approach: GTEx v8 Whole_Blood cis-eQTL API (singleTissueEqtl):
  b_e = nes (normalized effect), se_e = |nes|/Z (Z back-solved from pValue; algebraically consistent)
  Advantage: includes rsID + b38 positions; per-gene significant eQTLs fully retrievable by paging

## Design (final)
- Exposure: GTEx v8 Whole_Blood eQTL (n=670); outcome: FinnGen R11 GEST_DIABETES (16,802/237,816)
- Candidates: concordant(8) + reversed(8); macrosomia_only excluded because of n=1 artifacts
  (Y chromosome/CNV)
- Instruments: significant eQTLs, lead + distance pruning (>1Mb, max 5 SNPs); palindromic SNPs
  removed; strand harmonised
- Estimation: Wald (n=1) / IVW (n>1); OR per 1-SD genetically predicted expression

## Results (results/08_MR_results.csv)
- Genes with instruments: 8/16 (APOL6/FTH1/PAK3/RAB7B/PYCR1/KCNK12/SLC12A8/TMSB10 have no
  significant whole-blood eQTLs)
- Mostly single SNPs after distance pruning (GTEx significant eQTLs cluster around the lead)
- **IL1R2: OR=1.81 (95%CI 1.02-3.20), p=0.044** — higher genetically predicted IL1R2 expression →
  higher GDM risk
  (concordant class, consistent with the immune-activation direction shared by both arms; nominal
  significance only, does not pass 16-test Bonferroni correction at 0.0031)
- CRABP2: OR=0.93, p=0.088 (reversed class, marginally protective)
- Remaining 6 genes p>0.17

## Limitations (must appear in the paper's limitations)
1. GTEx is whole-blood eQTL, not placental tissue (GTEx has no placenta); expression–disease causal
   interpretation is limited by tissue relevance
2. No LD reference panel; distance pruning is an approximation; most IVs are single SNPs, so no
   pleiotropy tests (MR-Egger/leave-one-out)
3. Nominal significance does not survive multiple-testing correction → positioned as "suggestive
   evidence", not confirmation
4. FinnGen is a European population; cross-population eQTL extrapolation requires caution
