Improving causal effect estimation in multi-ancestry multivariable Mendelian randomization with transfer learning.

Area of Science:

• Genetics
• Epidemiology
• Statistical genetics

Background:

• Multivariable Mendelian randomization (MVMR) studies are crucial for causal inference but are often limited to European ancestry due to data availability.
• Existing MVMR methods lack the power to analyze underrepresented ancestries, hindering global health insights.

Purpose of the Study:

• To introduce MRBEE-TL, a novel multi-ancestry MVMR method.
• To enhance statistical power in underpowered ancestries for MVMR analyses.
• To enable the assessment of cross-ancestry heterogeneity in disease risk factor associations.

Main Methods:

• MRBEE-TL integrates transfer learning with bias-corrected estimating equations.
• The method leverages large European genome-wide association studies (GWAS) to improve power in other ancestries.
• It is designed to handle multi-ancestry data for robust causal inference.

Main Results:

• Simulations demonstrated that MRBEE-TL consistently outperformed existing MVMR methods.
• Real-world data analyses revealed MRBEE-TL's ability to identify ancestry-specific causal effects.
• The method significantly improved statistical power in African and East Asian ancestries.

Conclusions:

• MRBEE-TL overcomes limitations of traditional MVMR by enabling multi-ancestry analyses.
• This approach enhances the discovery of genetic associations across diverse populations.
• MRBEE-TL provides a powerful tool for global genetic epidemiology research.