Recalibrating Mendelian randomization under winner's curse, sample structure and polygenicity

Area of Science:

• Statistical Genetics
• Bioinformatics
• Epidemiology

Background:

• Mendelian randomization (MR) methods are widely used for causal inference.
• Existing MR methods suffer from bias and inflated type-I error rates in real data.
• Winner's curse and polygenicity-induced bias are key limitations.

Purpose of the Study:

• To develop a unified framework correcting for winner's curse and polygenicity in multivariable MR.
• To propose a novel multivariable Rao-Blackwellization approach.

Main Methods:

• Developed multivariable Rao-Blackwellization by conditioning on a noncentral chi-squared statistic.
• Derived closed-form conditional moments for bias correction.
• Integrated bias-corrected summary statistics into existing MR methods like MR-cML and MR-BEE.

Main Results:

• The proposed method effectively corrects for winner's curse while accounting for polygenicity and sample structure.
• Simulations and real data analyses demonstrate substantial improvement in type-I error control.
• Enhanced robustness of causal inference when combined with other MR techniques.

Conclusions:

• The novel multivariable Rao-Blackwellization offers a unified approach to address critical biases in MR.
• This method significantly improves the accuracy and reliability of genetic association studies.
• The approach is broadly applicable and enhances existing MR methodologies.