Improving on polygenic scores across complex traits using select and shrink with summary statistics (S4) and LDpred2

Area of Science:

• Genetics and Genomics
• Biostatistics
• Precision Medicine

Background:

• Polygenic scores (PGS) are crucial for clinical risk assessment in precision medicine.
• Developing accurate polygenic models is an ongoing challenge.
• The select and shrink with summary statistics (S4) method previously showed promise for epithelial ovarian cancer risk.

Purpose of the Study:

• To evaluate the performance of the S4 PGS method across 12 phenotypes in UK Biobank participants.
• To compare S4 PGS with the LDpred2 method and a combined S4+LDpred2 approach.
• To assess the accuracy of PGS models developed using only Genome-Wide Association Studies (GWAS) summary statistics.

Main Methods:

• Application of the S4 PGS method to 12 phenotypes in UK Biobank data.
• Comparison of S4 PGS with LDpred2 and a combined S4+LDpred2 method.
• Development of PGS models using only GWAS summary statistics to address data limitations.

Main Results:

• The S4+LDpred2 method demonstrated improved overall PGS accuracy across various phenotypes in UK Biobank participants.
• The S4+LDpred2 method achieved the highest estimated PGS accuracy in Finnish and Japanese populations.
• Successful development of PGS models solely from GWAS summary statistics was achieved.

Conclusions:

• The S4+LDpred2 method represents a significant advancement in polygenic score accuracy.
• This improved accuracy extends across multiple phenotypes and diverse populations.
• The method offers a robust approach for risk prediction using readily available GWAS summary statistics.