A robust method for genome-wide association meta-analysis with the application to circulating insulin-like growth factor I concentrations

Genome-wide association studies (GWAS) offer an excellent opportunity to identify the genetic variants underlying complex human diseases. Successful utilization of this approach requires a large sample size to identify single nucleotide polymorphisms (SNPs) with subtle effects. Meta-analysis is a cost-efficient means to achieve large sample size by combining data from multiple independent GWAS; however, results from studies performed on different populations can be variable due to various reasons, including varied linkage equilibrium structures as well as gene-gene and gene-environment interactions. Nevertheless, one should expect effects of the SNP are more similar between similar populations than those between populations with quite different genetic and environmental backgrounds. Prior information on populations of GWAS is often not considered in current meta-analysis methods, rendering such analyses less optimal for the detecting association. This article describes a test that improves meta-analysis to incorporate variable heterogeneity among populations. The proposed method is remarkably simple in computation and hence can be performed in a rapid fashion in the setting of GWAS. Simulation results demonstrate the validity and higher power of the proposed method over conventional methods in the presence of heterogeneity. As a demonstration, we applied the test to real GWAS data to identify SNPs associated with circulating insulin-like growth factor I concentrations.