Detecting signals in pharmacogenomic genome-wide association studies

Area of Science:

• Biostatistics
• Pharmacogenomics
• Genetics

Background:

• Pharmacogenomic studies often compare treatment outcomes across genetic subgroups.
• Identifying interactions between treatment and genotype is crucial for personalized medicine.
• Varying sample sizes across genotypes lead to unequal statistical power, complicating analysis.

Purpose of the Study:

• To address the statistical challenges of multiple testing and variable power in genome-wide pharmacogenomic analyses.
• To propose a robust statistical method for identifying genotype-specific treatment effects.
• To provide an easily implementable solution for analyzing pharmacogenomic data.

Main Methods:

• The study discusses the limitations of fixed P-value thresholds with variable statistical power.
• A solution based on Bayes factors is proposed to handle differing power across genetic subgroups.
• Priors are carefully specified, and methods are illustrated with clinical trial data.

Main Results:

• The proposed Bayes factor approach offers a more nuanced way to evaluate evidence for treatment-genotype interactions.
• It accounts for the implicit variation in Type I and Type II error costs due to unequal power.
• The method is demonstrated using a randomized controlled trial on folate supplementation and homocysteine levels.

Conclusions:

• A Bayes factor-based statistical method is presented for pharmacogenomic studies with variable power.
• This approach provides a more principled way to assess treatment effects across genetic subgroups.
• The method is implemented in R, offering practical utility for researchers.