A fully adjusted two-stage procedure for rank-normalization in genetic association studies

Area of Science:

• Genetics
• Statistical Genetics
• Bioinformatics

Background:

• Linear regression for genotype-phenotype associations is sensitive to non-normal trait distributions.
• Rarer variants are disproportionately affected by non-normality, impacting Type I error and statistical power.
• A common two-stage method involves rank-normalizing residuals, but this can introduce statistical issues.

Purpose of the Study:

• To identify and explain the statistical deficiencies of the widely used two-stage rank-normalization method in genetic association studies.
• To propose and evaluate an alternative, fully adjusted two-stage approach for improved statistical properties.

Main Methods:

• Theoretical analysis of the statistical properties of the standard and proposed two-stage methods.
• Application of both methods to genome-wide and whole-genome sequencing association study data.
• Evaluation of Type I error rates and statistical power.

Main Results:

• The standard two-stage method exhibits undesirable statistical properties due to a mis-specified mean-variance relationship and residual covariate associations.
• These issues lead to excess Type I errors and reduced statistical power, particularly for rare variants.
• The proposed fully adjusted two-stage approach effectively mitigates these problems.

Conclusions:

• The conventional two-stage rank-normalization method in genetic association studies is statistically flawed.
• A fully adjusted two-stage approach offers a more robust alternative, improving Type I error control and statistical power.
• This improved method is recommended for genotype-phenotype association analyses.