Genome-wide association studies (GWAS) have genotyped millions of SNPs on tens of thousands of individuals, providing a comprehensive investigation of genetic association with various traits including disease phenotypes, gene expression, and methylation. However, the power for genome-wide discovery is hindered by the need to account for multiple comparisons of testing millions of variants. In this talk, I will present the use of screening statistics to prioritize variants for testing, and cast this into an independent weighted hypothesis testing framework. The idea is to allocate the type I error to each hypothesis differently according to the screening statistic such that the power is maximized. A unique and ubiquitous feature of GWAS data is that genetic variants are correlated locally throughout the genome. Treating them as independent predictors yields conservative type I error control, thereby losing power. We extend the framework to account for these correlations, and show by simulation and a real data example that the proposed approach improves power and yield more findings than existing approaches.