Bayesian weighted Mendelian randomization for causal inference based on summary statistics.

Area of Science:

• Genetics and Bioinformatics
• Statistical Genetics
• Causal Inference

Background:

• Genome-Wide Association Studies (GWAS) offer opportunities for causal inference.
• Mendelian randomization (MR) is an instrumental variable (IV) method used with GWAS data.
• Polygenic traits and pleiotropy present unique challenges for MR compared to conventional IV methods.

Purpose of the Study:

• To propose a novel Bayesian weighted Mendelian randomization (BWMR) method for robust causal inference.
• To address challenges in MR, including polygenicity and pleiotropy.
• To develop a computationally efficient algorithm for BWMR.

Main Methods:

• Developed a Bayesian weighted Mendelian randomization (BWMR) model.
• Incorporated Bayesian weighting for outlier detection to handle pleiotropy.
• Utilized a variational expectation-maximization (VEM) algorithm for computational stability and efficiency.
• Derived a closed-form formula to correct posterior covariance estimation.

Main Results:

• BWMR demonstrated advantages over existing methods in comprehensive simulation studies.
• Applied BWMR for causal inference between 130 metabolites and 93 complex human traits.
• Identified novel causal relationships between exposure and outcome traits.

Conclusions:

• BWMR provides a robust and efficient approach for causal inference using GWAS data.
• The method effectively handles challenges posed by polygenicity and pleiotropy.
• BWMR facilitates the discovery of new causal links between biological traits and diseases.