MR-MAGIC: Robust Causal Inference Using Many Weak Genetic Interactions

2025-04-18

AI Summary

In Mendelian randomization (MR), invalidity of instrumental variables (IVs) is a pervasive source of bias in causal effect estimation. To address this, we propose the “Multiple Weak Genetic Interactions” (MWGI) identification framework, which leverages numerous weak but collectively informative gene–gene or gene–environment interactions as a novel source for causal identification—even when all candidate SNPs violate the IV assumptions (i.e., are entirely invalid). The method integrates high-dimensional weak-instrument asymptotics, genetic interaction modeling, robust moment estimation, and decomposition of exposure variation. Simulation studies and empirical analysis using UK Biobank data demonstrate that MWGI substantially improves robustness and statistical efficiency relative to conventional MR approaches. Crucially, it alleviates the strong reliance on IV validity, enabling consistent and asymptotically normal causal effect estimation under pervasive instrument invalidity. This work establishes a new paradigm for causal inference in complex disease epidemiology.

Technology Category

• Reasoning under Uncertainty: Causality
• Machine Learning: Causal Learning
• Multiagent Systems: Mechanism Design

Application Category

• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods
• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• User Modeling, Personalization and Recommendation: Explainable and interpretable methods for personalization

Abstract

Mendelian randomization (MR) studies commonly use genetic variants as instrumental variables to estimate causal effects of exposures on outcomes. However, the presence of invalid instruments-even when numerous-can lead to biased causal estimates. We propose a novel identification strategy that remains valid even when all candidate instruments are invalid by leveraging genetic interactions that collectively explain substantial exposure variation. Recognizing that individual interaction effects may be weak, we develop MR-MAGIC (Mendelian Randomization with MAny weak Genetic Interactions for Causality), a robust method that simultaneously addresses instrument invalidity and improves estimation efficiency. MR-MAGIC provides consistent and asymptotically normal estimates under a many-weak-interactions asymptotic framework. Comprehensive simulations and applications to UK Biobank data demonstrate that MR-MAGIC outperforms conventional MR methods in practice, offering reliable causal inference when standard approaches fail.

Problem

• Addresses bias from invalid instruments in Mendelian randomization
• Leverages weak genetic interactions for robust causal estimation
• Improves accuracy when standard MR methods fail

Innovation

• Leverages many weak genetic interactions
• Robust to invalid instrumental variables
• Improves estimation efficiency asymptotically