Constrained maximum likelihood-based Mendelian randomization robust to both correlated and uncorrelated pleiotropic...

Area of Science:

• Genetics and Bioinformatics
• Epidemiology
• Statistical Genetics

Background:

• Mendelian randomization (MR) uses genetic variants as instrumental variables (IVs) to investigate causal relationships from observational data.
• The validity of MR conclusions relies on assumptions that are often violated by pleiotropy, where IVs affect the outcome through pathways other than the exposure.
• Existing MR methods primarily address uncorrelated pleiotropy, with few capable of handling correlated pleiotropy, a significant challenge in real-world genetic studies.

Purpose of the Study:

• To develop and validate a novel MR method for robust causal inference from genome-wide association study (GWAS) summary data.
• To address the challenge of correlated pleiotropy, where genetic variants may be associated with unmeasured confounders.
• To improve the accuracy and power of causal inference in the presence of invalid instrumental variables and weak pleiotropic effects.

Main Methods:

• Proposed a constrained maximum likelihood and model averaging (cML-MA) approach for analyzing GWAS summary data.
• Enhanced the cML-MA method with data perturbation to manage situations with numerous invalid IVs and weak pleiotropic effects.
• Evaluated method performance through extensive simulations and real-world applications.

Main Results:

• The proposed cML-MA method demonstrated superior performance compared to existing MR approaches.
• Simulations confirmed that the new methods effectively control type I error rates and achieve higher statistical power.
• Applications to 48 risk factor-disease pairs, including cardio-metabolic diseases and asthma, validated the method's effectiveness.

Conclusions:

• The cML-MA approach offers a simple, effective, and robust solution for causal inference using GWAS summary data, particularly in the presence of correlated pleiotropy.
• The data perturbation enhancement further improves the method's reliability in challenging scenarios with invalid IVs.
• This work advances the application of MR for investigating complex trait and disease etiology using large-scale genetic data.