RBMR

Robust Bayesian Mendelian Randomization (RBMR): Bayesian framework for causal inference with linkage disequilibrium and pleiotropy

Robust Bayesian Mendelian Randomization (RBMR) implements a unified probabilistic framework for Mendelian randomization (MR) to estimate causal effects using genetic instrumental variables while accounting for linkage disequilibrium (LD), systematic pleiotropy, and idiosyncratic pleiotropy.

Key Features:

• Causal Inference via Genetic Instrumental Variables: Estimates causal effects of modifiable risk factors on outcomes using genetic variants as instrumental variables.
• Linkage Disequilibrium Modeling: Explicitly incorporates LD structure among correlated genetic variants to reduce bias and power loss.
• Idiosyncratic Pleiotropy Modeling: Uses a robust multivariate generalized t-distribution, represented as a Gaussian scaled mixture, to model heavy-tailed direct effects of genetic variants on outcomes with parameter estimation via EM-type algorithms.
• Statistical Inference Calibration: Computes standard errors by calibrating the evidence lower bound with a likelihood ratio test.

Scientific Applications:

• Genetic Epidemiology: Applied to benchmark datasets to identify associations, including coronary artery disease with increased risk of critically ill coronavirus disease 2019 (COVID-19).

Methodology:

RBMR integrates LD structure and pleiotropic effects within a Bayesian Mendelian randomization framework. It models direct genetic effects using a robust multivariate generalized t-distribution formulated as a Gaussian scaled mixture and performs parameter estimation with EM-type algorithms, with statistical validation based on calibrated evidence lower bounds and likelihood ratio testing.