Constructing Genetic Risk Scores: Robust Bayesian Approach through Projected Summary Statistics and Flexible...

Area of Science:

• Genetics
• Statistical genetics
• Computational biology

Background:

• Polygenic risk scores (PRS) leverage genome-wide association studies (GWAS) for disease risk stratification.
• Bayesian methods are favored for PRS due to model regularization and external information integration.
• Existing Bayesian PRS frameworks face potential issues with data source incompatibility.

Purpose of the Study:

• To advance Bayesian methods for developing more robust and accurate polygenic risk scores.
• To address the risk of posterior impropriety arising from integrating disparate GWAS and linkage disequilibrium (LD) data.
• To introduce a novel Bayesian PRS method with enhanced flexibility in modeling effect-size distributions.

Main Methods:

• Developed a projection method to ensure compatibility between GWAS summary statistics and LD data.
• Introduced a new PRS method, PRS-Bridge, utilizing a Bayesian bridge prior for flexible sparsity modeling.
• Conducted extensive benchmarking using synthetic and real datasets against alternative Bayesian PRS methods.

Main Results:

• The proposed projection technique resolves posterior impropriety issues in Bayesian PRS.
• PRS-Bridge demonstrates consistent and superior performance across diverse scenarios compared to existing methods.
• Prior specification and LD estimation strategies significantly impact PRS performance.

Conclusions:

• The novel projection technique and flexible Bayesian bridge prior offer a principled advancement for PRS development.
• PRS-Bridge provides a robust and high-performing tool for genetic risk prediction.
• These advancements have the potential to enhance clinical applications of PRS for disease risk stratification.