MR2G: A novel framework for causal network inference using GWAS summary data.

Area of Science:

• Genetics
• Systems Biology
• Statistical Genetics

Background:

• Causal network inference is crucial for understanding complex biological relationships and guiding interventions.
• Mendelian randomization (MR) uses genetic variants as instrumental variables (IVs) for causal inference, but struggles with unknown causal directions and cycles.
• Existing methods for network inference with IVs are limited, especially in handling feedback loops common in biological systems.

Purpose of the Study:

• To introduce MR2G, a novel statistical framework for robust causal network inference directly from GWAS summary statistics.
• To enable the inference of causal networks, including those with cycles (feedback loops), which are often challenging for standard MR approaches.
• To improve the accuracy and robustness of causal network inference in complex biological systems.

Main Methods:

• MR2G utilizes a formally defined recursive causal graph model linking direct causal effects to MR estimands.
• The framework recovers causal networks from pairwise MR effect estimates.
• It incorporates a network-informed IV screening strategy to mitigate pleiotropic bias and enhance robustness.

Main Results:

• Simulations demonstrate MR2G's superior accuracy and robustness in recovering complex causal structures, including feedback loops.
• Application to GWAS data for six complex diseases and nine cardiometabolic risk factors successfully identified known pathways.
• MR2G uncovered novel feedback relationships, showcasing its capability in disentangling intricate biological networks.

Conclusions:

• MR2G provides a robust method for inferring complex causal networks, including cycles, from large-scale genetic data.
• The framework enhances the understanding of biological relationships and disease etiology by revealing feedback mechanisms.
• MR2G is a valuable tool for dissecting complex causal networks relevant to human health and disease.