Incorporating functional annotation with bilevel continuous shrinkage for polygenic risk prediction

Area of Science:

• Genetics
• Statistical Genetics
• Bioinformatics

Background:

• Functional annotations can improve polygenic risk score (PRS) predictive performance.
• Bayesian shrinkage methods enhance PRS by accounting for sparse causal variants.
• Limited methods integrate both functional annotation and effect size shrinkage.

Purpose of the Study:

• Propose PRSbils, a novel PRS method using bilevel continuous shrinkage priors.
• Accommodate varying genetic architectures at both variant and functional annotation levels.
• Improve genetic risk prediction by integrating functional annotation information.

Main Methods:

• Developed PRSbils, a PRS method with a bilevel continuous shrinkage prior.
• Utilized functional annotation information (ANNOVAR, KEGG pathways).
• Applied the method to binary and quantitative traits in UK Biobank, MGI, and KoGES datasets.

Main Results:

• PRSbils achieved higher predictive performance (e.g., 8.0% higher AUC) than PRS-CS in simulations with variable group-wise heritability.
• Demonstrated improved AUC (average 6.4% higher) with overlapping annotation groups.
• Showcased PRSbils' potential for improving predictive performance in real-world data.

Conclusions:

• PRSbils effectively incorporates overlapping and non-overlapping annotations using bilevel shrinkage.
• The method enhances genetic risk prediction performance.
• Software is publicly available at https://github.com/styvon/PRSbils.