False Discovery Rate Control for High Dimensional Dependent Data with an Application to Large-Scale Genetic Association Studies