GWAS Components
FedGWAS supports the main blocks of a typical case–control GWAS screening lifecycle: federated quality control, KING-based relatedness screening, and logistic-regression association screening. Genetics operations are executed locally with PLINK; participating clients keep raw genotypes on site.
For stage ordering, see Workflow. For thresholds and chunk settings, see Parameters.
Supported screening components
| Component | Scope | What it screens / estimates |
|---|---|---|
| Federated QC (sample) | Per client | Sample missingness (--mind) |
| Federated QC (variant) | Cross-client | SNP missingness, MAF, Hardy–Weinberg equilibrium |
| Relatedness screening | Cross-client | Pairwise kinship via KING |
| Association screening (local filter) | Per client, then shared filter | Local logistic-regression screen using privacy-preserving tokens to drop jointly insignificant SNPs |
| Association screening (federated test) | Cross-client | Case–control logistic regression |
Federated quality control
- Sample QC: remove samples with high genotype missingness.
- Variant QC: remove SNPs that fail missingness, minor allele frequency (MAF), or Hardy–Weinberg equilibrium (HWE) thresholds agreed across clients.
Implementation stages are local_qc and global_qc / global_qc_response. See Quality Control.
Relatedness screening
- Estimate pairwise kinship with KING.
- Optionally filter related samples using
king_threshold.
Implementation stages are init_chunks and iterative_king. See KING / Kinship.
Association screening
- Local filter: each client runs local logistic regression and shares only privacy-preserving tokens for insignificant SNPs; clients then drop the shared insignificant set.
- Federated association screening: remaining variants are tested with case–control logistic regression across clients.
Current association screening is binary phenotypes via PLINK --logistic. Continuous traits, richer covariates, population stratification handling, and alternate association models are not covered yet.