Improved inference of polygenic effects using genome-wide association summary statistics

Genetic variants across the genome contribute to the architecture of complex traits, which are highly polygenic in nature. Understanding this complexity and improving genomic prediction requires accurate inference of the distribution of genetic effects. We introduce SumHEM, a summary-level heteroscedastic effects model that leverages summary statistics from genome-wide association studies (GWAS) to estimate genome-wide SNP effects efficiently. SumHEM outperformed state-of-the-art methods, including LDpred2, in heritability estimation, genetic effects inference, and genomic prediction through simulations and real data analyses. Notably, SumHEM demonstrated exceptional performance for highly polygenic traits, as evidenced by our analysis of 300 phenotypes from the UK Biobank. By providing a more precise genome-wide genetic effects profile, SumHEM offers a powerful approach for advancing polygenic risk prediction and trait mapping.