Benchmarking Mendelian randomization methods for causal inference using genome-wide association study summary...

Area of Science:

• Genetics
• Biostatistics
• Epidemiology

Background:

• Mendelian randomization (MR) is a popular causal inference method using genetic variants as instrumental variables (IVs).
• The reliability of MR methods in real-world applications with potential confounding remains uncertain.
• Existing research often relies on simulated data, limiting practical applicability.

Purpose of the Study:

• To benchmark the performance of 16 two-sample summary-level MR methods.
• To evaluate MR methods using real-world genetic datasets rather than simulations.
• To provide practical guidelines for selecting appropriate MR methods for causal inference.

Main Methods:

• Evaluation of 16 two-sample summary-level MR methods.
• Utilized real-world genetic datasets for a comprehensive benchmark study.
• Assessed type I error control, accuracy, replicability, and power across 1000 exposure-outcome trait pairs.

Main Results:

• Performance varied significantly across MR methods under different confounding scenarios (population stratification, pleiotropy, assortative mating).
• Identified specific methods with robust type I error control and accurate causal effect estimation.
• Highlighted differences in replicability and statistical power among the evaluated methods.

Conclusions:

• The choice of MR method significantly impacts causal inference reliability in real-world genetic data.
• Provides evidence-based recommendations for researchers to select appropriate MR methods.
• Advances best practices for applying MR in genetic epidemiology and related fields.