Bayesian Bidirectional Mendelian Randomization Under Correlated and Uncorrelated Pleiotropy Using GWAS Summary...

Area of Science:

• Biostatistics
• Statistical Genetics
• Epidemiology

Background:

• Mendelian randomization (MR) infers causality using genetic variants.
• Standard MR methods struggle with pleiotropy and bidirectional causality.
• Existing approaches often assume independent pleiotropic effects, which may not hold.

Purpose of the Study:

• Introduce BayBiMR, a Bayesian framework for bidirectional MR.
• Address limitations of unidirectional MR, including pleiotropy and feedback loops.
• Provide a robust tool for estimating causal effects in complex biological systems.

Main Methods:

• Developed a Bayesian bidirectional MR framework (BayBiMR).
• Incorporated hierarchical spike-and-slab priors for pleiotropy and Inverse-Wishart prior for SNP effect covariance.
• Utilized a small-feedback reduced-form approximation for efficient posterior computation via Gibbs sampling.
• Introduced BayBiMR(DP) for improved frequentist calibration in finite samples.

Main Results:

• BayBiMR and BayBiMR(DP) demonstrated superior type I error control compared to existing methods.
• The proposed methods showed particular strength in settings with correlated and mixed pleiotropy.
• Simulations confirmed improved performance in complex causal structures.
• Applications to real-world GWAS data highlighted practical utility.

Conclusions:

• BayBiMR offers a robust Bayesian approach for bidirectional causal inference.
• The framework effectively handles correlated and uncorrelated pleiotropy.
• BayBiMR serves as a valuable exploratory and sensitivity analysis tool for genetic epidemiology.