|
Abstract:
|
Accurate prediction of human disease risk can improve disease prevention and early-treatment. In order to reveal the associations between genetic variants and diseases, thousands of genome-wide association studies (GWAS) have been conducted. We can use data released from GWAS to build risk prediction model for human diseases. However, how to accurately identify potentially causal variants and estimate corresponding effect sizes are challenges.
In line with the empirical Bayes framework, we address these challenges by inferring genetic architecture, including the causal variant proportion and corresponding effect size distribution. Recent studies have showed that the causal variants are enriched in genomic/epigenomic functionally annotated regions. By integrating this domain knowledge, here we proposed an approach, namely GWEB-Anno, which estimates genetic architecture from GWAS summary statistics and diverse functional annotation data, and leverages the architecture information for genetic risk prediction. A merit of our approach is that there is no tuning parameter. Experiments showed our approach outperformed existing methods in terms of prediction accuracy.
|
