Non-parametric Polygenic Risk Prediction via Partitioned GWAS Summary Statistics

Area of Science:

• Complex trait genetics
• Genomic prediction
• Quantitative trait heritability

Background:

• Predicting phenotype from genotype is key to understanding trait heritability.
• Polygenic traits necessitate statistical methods combining numerous small genetic effects.
• Current polygenic risk prediction methods often require individual genotypes or explicit genetic architecture modeling.

Purpose of the Study:

• To propose a novel polygenic risk prediction method that bypasses the need for explicit genetic architecture modeling.
• To improve the accuracy and utility of polygenic risk scores for common diseases.
• To enable risk prediction using readily available genome-wide association study (GWAS) summary statistics.

Main Methods:

• Developed a non-parametric shrinkage (NPS) method using SNP effect sizes from GWAS summary statistics.
• Accounted for linkage disequilibrium (LD) by removing correlation structure in summary statistics.
• Applied piecewise linear interpolation on conditional mean effects.

Main Results:

• The NPS method reliably handles linkage disequilibrium across millions of dense genome-wide markers.
• NPS consistently improves polygenic risk prediction accuracy in both simulated and real datasets.
• Demonstrated improved identification of high-risk groups for breast cancer, type 2 diabetes, inflammatory bowel disease, and coronary heart disease.

Conclusions:

• The NPS method offers a powerful, non-parametric approach for polygenic risk prediction.
• This method enhances the practical application of genetic information for early disease intervention and prevention.
• NPS provides a valuable tool for identifying individuals at high risk for major complex diseases.