Risk factors affecting polygenic score performance across diverse cohorts

Area of Science:

• Genetics
• Biostatistics
• Population Health

Background:

• Polygenic scores (PGS) predict complex traits but their performance can vary.
• Factors beyond ancestry, such as personal and environmental covariates, may influence PGS accuracy.

Purpose of the Study:

• To analyze the impact of covariate stratification and interaction on body mass index polygenic score (PGS_BMI) performance.
• To identify covariates affecting PGS_BMI and explore methods to enhance model performance.

Main Methods:

• Analysis of PGS_BMI performance across European and African ancestry cohorts (N=512,723).
• Stratification by binary and continuous covariates, assessment of PGS_BMI-covariate interaction effects.
• Application of quantile regression, machine learning (neural networks), and GxAge GWAS for model improvement.

Main Results:

• 18/62 covariates showed significant, replicable R² differences in PGS_BMI performance.
• Covariates like age, sex, and lifestyle factors demonstrated substantial performance variations.
• Machine learning models improved relative R² by a mean of 23%, and GxAge GWAS by 7.8%.

Conclusions:

• Covariates significantly influence PGS_BMI performance and effects across diverse ancestries.
• Model performance can be enhanced by considering covariate interactions and nonlinear effects.
• Advanced modeling techniques offer improved accuracy for BMI prediction using PGS.