A flexible and unified framework for single- and multi-outcome Mendelian randomization using summary statistics.

Area of Science:

• Genetics and Bioinformatics
• Statistical Genetics
• Causal Inference

Background:

• Mendelian randomization (MR) is crucial for inferring causality between exposures and outcomes.
• Transcriptome-wide MR (TWMR) faces challenges like limited cis-quantitative trait loci (QTLs) and pervasive horizontal pleiotropy.
• Existing MR methods struggle with molecular traits and correlated outcomes.

Purpose of the Study:

• To introduce FusioMR, a robust Mendelian randomization framework for molecular and complex trait analyses.
• To develop single-outcome (FusioMRs) and multi-outcome (FusioMRm) models addressing specific MR challenges.
• To enhance the precision and power of causal inference in genetic studies.

Main Methods:

• FusioMRs utilizes gene-region-specific empirical priors and sampling-based inference for robustness with limited instruments.
• FusioMRm jointly analyzes correlated outcomes, leveraging shared instrumental variables (IVs) and pleiotropic effects.
• The framework integrates eQTL and genome-wide association study (GWAS) summary data.

Main Results:

• FusioMRs identified cell-type-specific gene expression traits associated with Alzheimer disease.
• FusioMRm detected alternative polyadenylation events linked to atrial fibrillation and ischemic stroke.
• FusioMRm improved estimation of low-density lipoprotein's causal effect on ischemic stroke by leveraging cross-ancestry data.

Conclusions:

• FusioMR provides a generalized and robust approach for Mendelian randomization in both molecular and complex trait studies.
• The framework enhances causal inference by addressing limitations in instrument availability and pleiotropy.
• FusioMR demonstrates broad applicability in genetic epidemiology and biomarker discovery.