A Litmus Test for Confounding in Polygenic Scores

Area of Science:

• Genetics
• Statistical Genetics
• Bioinformatics

Background:

• Polygenic scores (PGSs) are increasingly used for trait prediction.
• PGSs are assumed to capture direct genetic effects but are influenced by stratification, assortative mating, and dynastic effects (SAD).
• Understanding the contribution of SAD effects is crucial for accurate PGS interpretation and application.

Purpose of the Study:

• To develop and validate a method for partitioning the variance of PGSs into direct genetic effects and SAD effects.
• To quantify the relative contributions of direct and SAD effects on PGS variance across different genetic ancestries.
• To assess the impact of different genome-wide association study (GWAS) population structure adjustments on SAD variance.

Main Methods:

• Developed Partitioning Genetic Scores Using Siblings (PGSUS) method.
• Compared PGS derived from standard GWAS with PGS derived from sibling GWAS to isolate SAD effects.
• Analyzed variance components by axes of genetic ancestry to detect stratification and isotropic SAD variance.

Main Results:

• Demonstrated evidence of stratification in PGSs for height and educational attainment, and in UK Biobank PGSs.
• Observed ancestry-specific stratification of PGSs across different prediction samples and time periods (ancient vs. contemporary DNA).
• Showed that different population structure adjustment methods in GWAS have varying effectiveness in mitigating ancestry-specific and isotropic SAD variance.

Conclusions:

• PGSUS provides a nuanced interpretation of PGS variance by disentangling direct genetic effects from SAD effects.
• Family-based designs combined with population-based designs are essential for accurate interpretation and application of genomic predictors.
• Findings highlight the importance of accounting for SAD effects and ancestry when using PGSs in diverse populations.