Winner's Curse Free Robust Mendelian Randomization with Summary Data.

Area of Science:

• Genetics
• Epidemiology
• Biostatistics

Background:

• Mendelian Randomization (MR) uses genetic variants for causal inference from genome-wide association studies (GWAS) summary data.
• Classical MR methods are susceptible to biases from the winner's curse and pleiotropy.
• Existing robust MR approaches have limitations in addressing these biases effectively.

Purpose of the Study:

• To develop a unified robust Mendelian Randomization framework for causal inference using GWAS summary data.
• To systematically address and mitigate biases caused by the winner's curse and pleiotropy.
• To enable valid statistical inference on causal effects without stringent distributional assumptions on pleiotropic effects.

Main Methods:

• A novel robust Mendelian Randomization framework is proposed.
• The framework systematically removes winner's curse bias.
• It screens out genetic instruments exhibiting pleiotropic effects.

Main Results:

• The proposed framework provides valid statistical inference for causal effects.
• The estimator demonstrates convergence to a normal distribution with well-estimable variance under appropriate conditions.
• Performance was validated through Monte Carlo simulations and two case studies.

Conclusions:

• The developed MR framework offers a robust approach to causal inference from summary data.
• It effectively addresses winner's curse and pleiotropy, improving the reliability of genetic epidemiology findings.
• An R package, MRcare, is available for practical application of the method.