Model checking via testing for direct effects in Mendelian Randomization and transcriptome-wide association studies.

Area of Science:

• Genetics
• Statistical Genetics
• Epidemiology

Background:

• Causal inference in observational studies often relies on genetic variants as instrumental variables (IVs) to mitigate confounding.
• Mendelian randomization (MR) and transcriptome-wide association studies (TWAS) are popular methods but are sensitive to violations of IV assumptions, such as horizontal pleiotropy.
• Existing methods for assumption checking are limited, especially for TWAS, and can yield conflicting results.

Purpose of the Study:

• To develop a general goodness-of-fit (GOF) test for assessing the validity of assumptions in both MR and TWAS.
• To address the challenge of horizontal pleiotropy and other assumption violations that bias causal inference.
• To provide a reliable tool for practitioners to check model assumptions and interpret results in genetic epidemiology.

Main Methods:

• Introduction of TEDE (TEsting Direct Effects), a novel GOF test.
• TEDE is designed to be applicable to both independent SNPs (common in MR) and correlated SNPs (common in TWAS).
• Validation through simulation studies and real-world data examples.

Main Results:

• The proposed TEDE test demonstrates high statistical power.
• Simulation studies and real data analyses confirm frequent violations of modeling and IV assumptions in practice.
• TEDE offers advantages over existing methods and highlights the importance of model checking.

Conclusions:

• TEDE is a powerful and generalizable tool for assessing the validity of assumptions in MR and TWAS.
• The frequent violation of assumptions underscores the necessity of rigorous model checking in genetic causal inference.
• Implementing TEDE can improve the reliability and interpretability of causal conclusions drawn from observational data.