Inferring causal direction between two traits using R² with application to transcriptome-wide association studies.

Area of Science:

• Genetics
• Statistical Genetics
• Bioinformatics

Background:

• Mendelian randomization (MR) uses genetic variants as instrumental variables (IVs) to infer causality.
• Existing MR methods like MR Steiger's and CD-Ratio rely on single nucleotide polymorphisms (SNPs) to determine causal direction.
• Transcriptome-wide association studies (TWASs) often face challenges with small sample sizes for gene expression data.

Purpose of the Study:

• To develop a novel R-squared based method for combining information from multiple SNPs to infer causal direction.
• To generalize existing single-SNP MR methods to a multi-SNP framework.
• To enhance causal inference in TWASs and similar applications, especially with limited sample sizes.

Main Methods:

• Proposed a new method utilizing the coefficient of determination (R-squared) to integrate information from multiple, potentially correlated SNPs.
• Extended Steiger's method to accommodate multiple SNPs as IVs.
• Introduced R2S, a novel approach for selecting and removing invalid IVs to improve robustness.
• Applied the method to individual-level GTEx gene expression and UK Biobank GWAS data, as well as GWAS summary data.

Main Results:

• The proposed multi-SNP R-squared method demonstrated advantages over existing methods in simulations.
• Successfully identified known and novel causal genes for high/low-density lipoprotein cholesterol (HDL/LDL).
• Applied to GWAS summary data, the method inferred causal relationships between HDL/LDL and stroke/coronary artery disease (CAD).

Conclusions:

• The new R-squared based multi-SNP method offers a more flexible and powerful approach for causal inference in genetic studies.
• The method enhances robustness by incorporating a strategy for invalid IV detection (R2S).
• The findings have implications for understanding gene-trait relationships and disease etiology, particularly in TWASs.