Multi-ancestry Meta-analysis and LD Score Regression

Although large strides have been made in genome-wide association studies (GWASs) and polygenic prediction for individuals of European ancestry, these activities have lagged behind for non-European samples. Unfortunately,... [ view full abstract ]

Although large strides have been made in genome-wide association studies (GWASs) and polygenic prediction for individuals of European ancestry, these activities have lagged behind for non-European samples. Unfortunately, existing methods for meta-analysis do not account for cross-ancestry differences in genetic architecture and linkage disequilibrium (LD), making it impossible to share information from existing, European-based GWAS summary statistics to GWAS summary statistics from cohorts of non-European descent. In this paper, we develop a principled method, Multi-Ancestry Meta-Analysis (MAMA), that meta-analyzes GWAS summary statistics based on samples of different ancestries accounting for differences in LD pattern and genetic architecture. This method increases the precision of GWAS effect-size estimates for both ancestries, improving polygenic prediction and elucidating biological pathways. A key component of MAMA is an extension of LD score regression which accounts for LD differences across populations. Using our extension of LD score regression, we can estimate the genetic correlation of phenotypes measured in different ancestry groups. We illustrate the improved precision of effect size estimates and increase predictive accuracy of polygenic scores when using MAMA in simulation and in applications to a number of anthropometric and psychiatric phenotypes.