Sequence Robust Association Test (SRAT)

A Rank-Based Method for Testing Genetic Associations with Familial Data

Source: vignettes/SRAT.Rmd

Introduction

The Sequence Robust Association Test (SRAT) is a fully rank-based, flexible approach to test for association between a set of genetic variants and an outcome, while accounting for within-family correlation and adjusting for covariates. SRAT was developed specifically for genetic association studies that involve family data, such as genome-wide association studies (GWAS) and next-generation sequencing studies (NGSS).

Key Features and Advantages

SRAT offers several advantages over traditional methods:

  1. Robustness: As a fully rank-based method, SRAT is robust to outliers and non-normality in the outcome distribution.

  2. Flexibility in correlation structure: SRAT allows for unknown correlation structures within families.

  3. Improved type I error control: SRAT provides better protection against type I error rate inflation compared to existing methods.

  4. Enhanced power: For settings with skewed outcome distributions, SRAT can be more powerful than parametric approaches.

  5. Special case: SRAT includes the well-known Wilcoxon rank sum test as a special case.

Basic Usage

Setup

First, load the required packages:

library(SRAT)
library(SKAT)  # For comparison and example data

Example Data

For demonstration, we’ll use the example data from the SKAT package, which contains: - Z: A genotype matrix containing genetic variants - X: A covariate matrix (e.g., age, gender) - y.c: A continuous phenotype variable

data(SKAT.example)
attach(SKAT.example)

Comparing SKAT and SRAT

Let’s first run the analysis using SKAT for comparison:

# Run SKAT analysis
obj.skat <- SKAT_Null_Model(y.c ~ X, out_type = "C")
skat_pval <- SKAT(Z, obj.skat, weights.beta = c(1, 1))$p.value
print(paste("SKAT p-value:", skat_pval))

Now, let’s run the same analysis using SRAT:

# Run SRAT analysis
set.seed(123)  # For reproducibility
obj.srat <- srat.null(y.c, X)  # Fit null model
srat_pval <- srat.test(Z, obj.srat)  # Test for association
print(paste("SRAT p-value:", srat_pval))

SRAT Workflow Explained

SRAT operates in a two-step process:

  1. Null model fitting (srat.null()): This function fits a null model that accounts for the covariates while ignoring the genetic variants. It returns an object containing elements needed for the subsequent testing.

  2. Association testing (srat.test()): This function tests for the association between the genetic variants and the outcome, conditional on the null model.

Parameters and Options

srat.null()

  • y: Outcome vector (e.g., phenotype)
  • X: Covariates for adjustment (e.g., age, gender)
  • B: Number of perturbations (default: 1000)

srat.test()

  • Z: Covariates for testing (e.g., genotype)
  • obj: Object from null model
  • w.sqrt: Square root of weight vector (default: all 1’s)

Advanced Example: Using Weights

SRAT allows for weighting of genetic variants, which can improve power when certain variants are more likely to be causal:

# Create weights (example: upweight rare variants)
weights <- rep(1, ncol(Z))
maf <- colMeans(Z)/2
rare_variants <- which(maf < 0.05)
weights[rare_variants] <- 2

# Apply weights in SRAT
obj.srat <- srat.null(y.c, X)
weighted_result <- srat.test(Z, obj.srat, w.sqrt = sqrt(weights))
print(paste("SRAT p-value with weights:", weighted_result))

Understanding the Output

The p-value returned by SRAT indicates the statistical significance of the association between the set of genetic variants and the phenotype. A smaller p-value suggests stronger evidence against the null hypothesis of no association.

Comparison with Other Methods

SRAT has been shown to be particularly advantageous when: - The outcome distribution is skewed or contains outliers - Family correlation structures are complex - The sample size is relatively small

Conclusion

SRAT provides a robust alternative to existing methods for testing genetic associations in family-based studies. Its rank-based nature and ability to handle family correlation make it suitable for a wide range of genetic studies.

References

Dai W, Yang M, Wang C, Cai T. Sequence robust association test for familial data. Biometrics. 2017 Sep;73(3):876-884. doi: 10.1111/biom.12643.