Mendelian randomization analysis using mixture models for robust and efficient estimation of causal effects.

Area of Science:

• Genetics
• Epidemiology
• Biostatistics

Background:

• Mendelian randomization (MR) is a key method for inferring causality using genetic variants.
• Horizontal pleiotropy poses a significant challenge, potentially biasing MR results.

Purpose of the Study:

• To introduce MRMix, a novel, robust, and efficient MR analysis method.
• To address bias from horizontal pleiotropy in MR analyses with numerous genetic instruments.

Main Methods:

• Development of a novel spike-detection algorithm.
• Application of a normal-mixture model for effect-size distributions.
• Utilizing large numbers of genetic instruments for enhanced MR analysis.

Main Results:

• MRMix demonstrates nearly unbiased or less biased causal effect estimates compared to alternatives.
• The method achieves higher efficiency than other robust estimators.
• Identified causal links: BMI and age-at-menarche (breast cancer), BMI (major depressive disorder).
• Found no causal effect: HDL and triglycerides (coronary artery disease).

Conclusions:

• MRMix offers a robust and efficient approach for Mendelian randomization analysis.
• The method effectively mitigates bias from horizontal pleiotropy.
• New insights into causal relationships between various traits and diseases were obtained.