MUSSEL: Enhanced Bayesian polygenic risk prediction leveraging information across multiple ancestry groups

Area of Science:

• Genetics
• Statistical genetics
• Population genetics

Background:

• Polygenic risk scores (PRSs) predict complex traits but show performance disparities across ancestral populations.
• Existing methods struggle to bridge the prediction gap in diverse ancestries.

Purpose of the Study:

• To develop and evaluate MUSSEL, a novel method for ancestry-specific polygenic prediction.
• To improve PRS accuracy in underrepresented populations by integrating multi-ancestry GWAS summary statistics.

Main Methods:

• MUSSEL employs Bayesian hierarchical modeling and ensemble learning to borrow information across ancestry groups.
• The method utilizes summary statistics from genome-wide association studies (GWASs).
• Evaluated through simulations and analyses of four large studies with diverse participants (5.7 million total).

Main Results:

• MUSSEL demonstrated substantial prediction R-squared gains compared to existing methods (e.g., PRS-CSx, CT-SLEB) in African ancestry populations.
• Average gains of 40.2% and 49.3% were observed for 11 continuous traits.
• Optimal method performance varied based on GWAS sample size, target ancestry, trait architecture, and LD reference panels.

Conclusions:

• MUSSEL offers a promising approach for enhancing PRS accuracy in diverse populations.
• A combination of methods may be necessary for generating the most robust PRSs across all ancestries.
• Further development is needed to optimize PRS for global applicability.