QC Test Set - Methods

GWAS method: REGENIE with dosages (regenie_dosages)
Visualization: gwas_viz_combined.R (Manhattan plots)

Contingency QC:
- Binary phenotypes: MIN_CELL = min(carrier_case, carrier_ctrl)
  Only alternative allele carriers are considered (reference allele excluded).
- Continuous phenotypes: MIN_CELL = total carrier count
  All samples are treated as "cases" (no case/control split).
- Variants with MIN_CELL <= 3 are flagged with black rings on Manhattan plots.

Recommended filtering thresholds (based on test set analysis):
- Binary: MIN_CELL >= 5 (removes 93% of LOF/CNV artifacts, keeps 99% of SNV/CYP signals)
- Continuous: MIN_CELL >= 10 (higher threshold for total carrier count)
- No separate treatment needed for dosage vs hardcall genotypes

Threshold impact (tested on positive/negative controls):
  simvastatin__prescribed (POSITIVE): 294/296 sig variants survive at MC>=5 (99.3%)
  warfarin__prescribed (POSITIVE): 94/131 sig survive (72%; removed are mostly CNV)
  zopiclone__M796 (NEGATIVE): 1/14 sig survive (93% removed - LOF artifacts)
  propranolol__prescribed (NEGATIVE): 32/46 sig survive (30% removed - CNV/LOF)

What gets removed (MIN_CELL < 5):
- LOF: carrier_case=1, extreme BETA (20-60). Always artifacts in ADR phenotypes.
- CNV regions: carrier_case=2-4, moderate BETA. Mostly artifacts.
- SNV: almost never removed (MIN_CELL typically > 100).
- Notable edge case: PROS1_lof in warfarin (genuine gene, MIN_CELL=4) is borderline.

Test set: 38 phenotypes (15 negative, 21 positive, 2 mixed)
- negative: top variants are artifacts, should be removed by filtering
- positive: top variants are genuine signals, should survive filtering
- mixed: unclear or mixed results

Phenotype types:
- binary: case/control (ADR ICD codes, prescribed yes/no, diagnoses)
- continuous: quantitative trait (optimal dose, biomarkers, measurements)

Variant classes:
- SNV: single nucleotide variants (chr1-22, X)
- CYP: CYP pharmacogene variants (cypmicro hardcall + cypdosage dosage)
- HLA: HLA variants (chr6)
- LOF: loss-of-function variants (WES)
- MPC: missense variants with MPC scores (WES, dosage)
- CNV: copy number variants (genes + regions)