Comparison of Methods Utilizing Sex-Specific PRSs Derived From GWAS Summary Statistics

Area of Science:

• Genetics
• Statistical Genetics
• Bioinformatics

Background:

• Polygenic risk scores (PRS) estimate genetic susceptibility but often ignore known sex differences in complex traits.
• Sex-specific genetic architectures for many traits remain poorly understood, complicating PRS development.
• Current PRS methods typically use combined-sex Genome-Wide Association Studies (GWAS) data, potentially reducing accuracy for sex-differentiated conditions.

Purpose of the Study:

• To benchmark the performance of sex-specific versus sex-agnostic polygenic risk scores (PRS) for traits exhibiting sex differences.
• To evaluate different PRS models (single-population and multiple-population) using both simulated and real-world genetic data.
• To provide guidance on optimizing PRS calculation strategies as GWAS sample sizes increase.

Main Methods:

• Simulated genetic data with varying genetic correlations and sample size ratios between sexes were used.
• Two PRS models, single-population (PRScs, LDpred2) and multiple-population (PRScsx), were applied to sex-specific and sex-agnostic GWAS data.
• Real data from 19 traits in the GIANT consortium and UK Biobank were analyzed using both sex-specific and sex-agnostic GWAS summary statistics.

Main Results:

• Sex-specific PRS combining LDpred2 or PRScsx models showed highest prediction accuracy under low genetic correlation and balanced, large sample sizes.
• Sex-agnostic PRS (LDpred2, PRScs) performed better when genetic correlation was high or sample sizes were unbalanced/small.
• For real-world data, incorporating sex-specific information improved PRS prediction accuracy for waist-to-hip ratio (WHR) related traits.

Conclusions:

• Combining sex-specific PRSs can outperform sex-agnostic approaches for sex-differentiated traits, particularly with sufficient and balanced data.
• The optimal PRS strategy depends on genetic correlation between sexes, sample size balance, and the chosen PRS model.
• These findings offer practical recommendations for developing more accurate PRSs for sex-differentiated traits in future large-scale GWAS.