An integrative multi-context Mendelian randomization method for identifying risk genes across human tissues.

Area of Science:

• Genetics and Bioinformatics
• Statistical Genetics
• Causal Inference

Background:

• Conventional Mendelian randomization (MR) methods struggle with mapping genetic risk factors due to limitations in expression quantitative trait loci (eQTLs) as instrumental variables (IVs).
• Tissue-specific eQTL effects violate the core MR assumption of consistent IV effects across different data contexts (e.g., eQTL and GWAS data).
• Estimating sparse causal effects and identifying relevant molecular traits for complex diseases remains a significant challenge in genetic epidemiology.

Approach:

• We introduce mintMR, a novel multi-context multivariable integrative MR framework designed to address the limitations of conventional MR methods.
• mintMR models molecular trait effects across multiple tissues and gene regions simultaneously, utilizing eQTLs with consistent effects across tissues to enhance IV consistency.
• The framework employs multi-view learning to integrate information across tissues, molecular traits, and gene regions, iteratively refining disease relevance probabilities and improving sparse effect estimation.

Key Points:

• mintMR enhances the reliability of instrumental variables by selecting eQTLs with consistent effects across multiple tissue types.
• The integration of multi-view learning allows for a collective modeling of disease relevance indicators, capturing complex patterns across diverse biological data.
• The iterative approach of mintMR improves the estimation of sparse causal effects, particularly crucial for identifying subtle genetic influences on complex diseases.

Conclusions:

• The proposed mintMR framework effectively addresses challenges in MR for gene mapping, offering improved causal effect estimation.
• Application of mintMR to 35 complex traits using multi-tissue QTLs demonstrates its ability to control for genome-wide inflation.
• mintMR provides novel insights into disease mechanisms by integrating multi-context molecular trait data within a robust causal inference framework.