sPLINK

sPLINK implements a federated learning framework for conducting privacy-preserving genome-wide association studies (GWAS) across distributed cohorts.

Key Features:

• Federated Learning Approach: Employs a federated learning framework that integrates distributed genotype and phenotype datasets without sharing raw data.
• Privacy Preservation: Performs decentralized model training where only model parameters are communicated between cohorts and a central server, preventing raw data exchange and supporting compliance with regulations such as GDPR.
• Accuracy and Robustness: Produces results equivalent to aggregated analyses performed with PLINK and remains robust to cross-cohort heterogeneity in phenotype and confounding factors.
• Statistical Tests and Efficiency: Supports chi-square and linear/logistic regression tests with practical runtimes (minutes to hours) and controlled network bandwidth consumption for large-scale studies.

Scientific Applications:

• Collaborative GWAS: Enables multi-institution GWAS without pooling raw data, facilitating larger effective sample sizes for association discovery.
• Alternative to Meta-analysis: Provides a federated alternative to traditional meta-analysis that mitigates accuracy loss due to cross-study heterogeneity.
• Complex Disease Association Discovery: Supports large-scale genetic research aimed at identifying genetic variants associated with diseases and improving precision of genetic predictors.

Methodology:

Implements federated learning with decentralized model training and parameter exchange, performs chi-square and linear/logistic regression tests, and validates results against aggregated PLINK analyses.