T1: Incorporate DougSpeed’s code into a wrapper function [Srikant/Enamul]

[espenhgn]

Generate 10 different versions of data with different sample size, etc.

• Phenotype
• Genotype
• Covariates

[Enamul-Hoq]

T1 Plan: DougSpeed Wrapper + Synthetic Dataset Variants

I will take care of T1 (DougSpeed wrapper) together with Srikant.

Goal:
Create a reproducible wrapper that generates 10 synthetic dataset variants, each consisting of 10 sites, with outputs suitable for federated GWAS / FL experiments.

---

Deliverables (per dataset variant v01–v10)

For each site (site01–site10):

• genotypes.bed / genotypes.bim / genotypes.fam (PLINK, hg38 assumption)
• phenotype.pheno (FID IID PHENO) and/or phenotype in .fam column 6
• covariates.cov (FID IID age sex PC1…PCk)

At dataset root:

• dataset_config.json (all parameters used)
• (optional but recommended) truth_causal.csv (causal SNPs + effect sizes)

---

Proposed implementation

• synthgen/wrapper.py: generate_dataset_variant(cfg, out_dir)
• configs/variants.yaml: defines the 10 variants (sample size, site imbalance, prevalence, effect size strength, polygenicity)
• Run via:

  python -m synthgen.run --config configs/variants.yaml --out data_synth/