The primal goal of genome-wide association studies (GWAS) is to discover genetic variants associated with diseases and other phenotypes. In addition, information from GWAS data can be used to accomplish other objectives, such as estimating the heritability of the phenotypes, inferring potential causal variants and predicting the genetic risks of the phenotypes for individuals. Although these tasks are related, existing methods have been mostly developed for these objectives separately. We note that all the objectives stated above are related with the global information among genetic variants, such as the proportion of causal variants and the distribution of their effect sizes. Here, under the general empirical Bayes framework, we have developed an approach, namely GWEB, which infers the global information from GWAS summary statistics, and leverages such information for heritability estimation, statistical fine-mapping and genetic risk prediction simultaneously under a unified framework. Simulation and empirical experiments demonstrated that our approach can achieve better performance than state-of-the-arts methods for each individual objective.