HERE we are maintaining an newly improved stable version of SAIGE and SAIGE-GENE+. Please find the https://saigegit.github.io/SAIGE-doc/ for documentation.

SAIGE is an R package developed with Rcpp for genome-wide association tests in large-scale data sets and biobanks. The method

• accounts for sample relatedness based on the generalized mixed models
• allows for model fitting with either full or sparse genetic relationship matrix (GRM)
• works for quantitative and binary traits
• handles case-control imbalance of binary traits
• computationally efficient for large data sets
• performs single-variant association tests
• provides effect size estimation through Firth's Bias-Reduced Logistic Regression
• performs conditional association analysis

SAIGE-GENE (now known as SAIGE-GENE+) are new method extension in the R package for testing rare variant in set-based tests.

• performs BURDEN, SKAT, and SKAT-O tests
• allows for tests on multiple minor allele frequencies cutoffs and functional annotations
• allows for specifying weights for markers in the set-based tests
• performs conditional analysis to identify associations independent from nearly GWAS signals

The package takes genotype file input in the following formats

• PLINK (bed, bim, fam), PGEN, BGEN, VCF, BCF, SAV

saige-rs: Rust Implementation

saige-rs is a Rust rewrite of the SAIGE GWAS tool located in the saige-rs/ directory. It reimplements the core statistical algorithms (GLMM fitting, SPA, score tests, variance ratio estimation) with native performance, memory-mapped I/O, and rayon parallelism.

Building saige-rs

cd saige-rs
cargo build --release
# Binary: target/release/saige

Usage

# Step 1: Fit null GLMM
saige fit-null \
  --plink-file <prefix> \
  --pheno-file <pheno.txt> \
  --pheno-col <trait> \
  --covar-cols <cov1,cov2> \
  --trait-type binary \
  --output-prefix <output>

# Step 2: Association tests
saige test \
  --model-file <output.saige.model> \
  --bgen-file <file.bgen> \
  --output-file <results.txt> \
  --is-spa --is-fast-spa

Benchmarks: saige-rs vs R SAIGE

Simulated data: 2,000 samples, 5K GRM markers, 2,000 test variants, binary trait (~10% prevalence).

Step	R SAIGE	saige-rs	Speedup
Step 2: Association tests	5.2s	0.12s	43x

R SAIGE run via Docker (linux/amd64). saige-rs native arm64 on Apple Silicon.

P-value concordance

Both implementations produce highly concordant p-values across 1,998 tested variants with SPA correction enabled.

Metric	Value
P-value correlation (-log₁₀)	0.995
BETA correlation	0.9997
SE correlation	0.9994
Sign concordance	99.4%

Note: R SAIGE must be run with --is_fastTest=FALSE for accurate variance estimation. The default --is_fastTest=TRUE uses an approximate formula that can produce inflated variance for high-AF variants.