|
Abstract:
|
Polygenic risk prediction in diverse populations currently falls far behind risk prediction in populations of European descent. The use of European training samples for risk prediction in non-European populations reduces prediction accuracy due to different patterns of linkage disequilibrium (LD). Here, we propose a novel penalized regression based polygenic risk prediction method for cross-ethnic studies. We introduce a new cross-ethnic penalty function to incorporate different LD structure across multiple populations and evaluate its predictive performance with a sparsity penalty such as the Lasso and the minimax concave penalty (MCP). This function can improve the prediction accuracy for complex traits of non-European ancestry (primary ethnicity) using European ancestry (secondary ethnicity) with large samples. We performed large-scale simulation studies to illustrate the excelling performance of our multi-ethnic approach compared to single-ethnic one. We further applied the proposed method to real GWAS data sets with 1000 Genome imputation from UK Biobank (N=456,898) and Biobank Japan (N=180,868).
|
