Power analysis of transcriptome-wide association study: Implications for practical protocol choice

Area of Science:

• Genetics
• Statistical Genetics
• Bioinformatics

Background:

• Genome-wide association studies (GWAS) identify genetic variants associated with phenotypes.
• Transcriptome-wide association studies (TWAS) integrate gene expression data to link variants to phenotypes through mediating gene expression.
• The predictive accuracy of genotype-expression models in TWAS is often low, impacting study power.

Purpose of the Study:

• To investigate the impact of expression model accuracy on TWAS power.
• To compare the statistical power of GWAS, TWAS, and a hypothetical method using real expression data.
• To determine the conditions under which TWAS is advantageous over GWAS.

Main Methods:

• Derived non-centrality parameters (NCPs) for linear mixed models (LMMs) to calculate statistical power.
• Compared power across GWAS, TWAS (using predicted expression), and a hypothetical protocol (using real expression).
• Examined scenarios of causality and pleiotropy, and varied expression heritability.

Main Results:

• Under pleiotropy, TWAS with predicted expression outperformed TWAS with actual expression, validating TWAS utility even with imperfect models.
• GWAS surpassed TWAS when expression heritability fell below 0.04 (causality) or 0.06 (pleiotropy).
• Evidence suggests TWAS may have been inappropriately used in studies with low expression heritability.

Conclusions:

• TWAS remains a valuable tool, particularly under pleiotropy, even with low predictive accuracy for gene expression.
• The choice between GWAS and TWAS depends critically on expression heritability and genetic architecture (causality vs. pleiotropy).
• Careful consideration of expression heritability is crucial to avoid misapplication of TWAS in genetic association studies.