Variant selection to maximize variance explained in cis-Mendelian randomization.

Area of Science:

• Genetics
• Statistical Genetics
• Epidemiology

Background:

• Cis-Mendelian randomization (MR) relies on instrumental variables (IVs) from a single gene region.
• High linkage disequilibrium (LD) among variants complicates optimal IV selection.
• Using only the lead variant may limit statistical power when multiple causal signals exist.

Purpose of the Study:

• To compare methods for selecting IVs that incorporate correlated non-lead variants.
• To evaluate the ability of these methods to increase instrument strength (variance explained, R²).
• To assess improvements relative to the lead-variant-only approach in cis-MR.

Main Methods:

• Compared LD-pruning, conditional and joint analysis (COJO), Sum of Single Effects (SuSiE) regression, and principal component analysis (PCA).
• Applied methods to the haptoglobin (HP) gene region, simulated traits, and 15 additional gene regions.
• Estimated R² using variant-protein association estimates (Fenland study) and LD data (UK Biobank).

Main Results:

• Four methods showed a median 145.1% proportional gain in R² in the HP region compared to the lead variant alone.
• These methods achieved a median 36.3% reduction in MR standard error.
• All methods successfully recovered expected genetic variance in simulations and outperformed the lead-variant-only approach in gene region analyses.

Conclusions:

• Methods incorporating correlated non-lead variants reliably enhance instrument strength in cis-MR.
• These approaches offer a significant improvement over using only the lead variant.
• Recommend using these methods while comparing with lead-variant-only estimates to ensure stability.