SAIGEgds

SAIGEgds performs high-throughput phenome-wide association studies (PheWAS) using generalized linear mixed models to test single-variant associations in large-scale genotype datasets.

Key Features:

• Generalized linear mixed models (GLMM): Implements GLMMs to model phenotypes while accounting for relatedness and population structure.
• SPAtest in C with saddlepoint approximations: Includes SPAtest functions implemented in C to compute saddlepoint approximations for score tests.
• Optimized C++ codebase: Uses an optimized C++ implementation to improve computational performance over the original R implementation.
• Genomic Data Structure (GDS) integration: Reads and processes genotype data from GDS files.
• Sparse genotype dosage support: Handles sparse genotype dosages efficiently for imputed whole-genome data.
• Single-variant testing at biobank scale: Performs single-variant association tests across millions of variants and hundreds of thousands of samples.
• Control of sample structure and case-control imbalance: Incorporates methods to control for sample structure and case-control imbalance in association tests.
• Performance improvement versus SAIGE R v0.29.4.4: Demonstrates approximately 5–6× speedup on UK Biobank White British genotype data (~430,000 samples) compared to the original SAIGE R package.
• High-performance computing suitability: Designed for execution on high-performance computing clusters for biobank-scale analyses.

Scientific Applications:

• Phenome-wide association studies (PheWAS): Large-scale testing of genotype–phenotype associations across many phenotypes.
• ICD10-coded phenotype association analysis: Association analysis between thousands of ICD10-coded phenotypes and genotypes.
• Biobank-scale single-variant association testing: Single-variant association testing across millions of variants in cohorts such as the UK Biobank.
• Analysis of whole-genome imputed genotype data: Use with imputed genotype dosages for genome-wide association analyses.

Methodology:

Implements generalized linear mixed models; integrates SPAtest functions in C to compute saddlepoint approximations; uses an optimized C++ implementation; reads Genomic Data Structure (GDS) files and handles sparse genotype dosages to perform single-variant tests; builds on the SAIGE R package v0.29.4.4.