Genome-wide association studies (GWASs) are aimed to identify genomic loci that are associated with common and complex traits, such as auto-immune diseases and psychiatric disorders. Strong evidence suggests that many common diseases are genetically heterogeneity, which is unobserved in most GWASs and lead to power loss of the common used association tests. Therefore, we proposed a likelihood ratio test (gLRTH) in the present of latent population heterogeneity and then conducted extensive simulation studies to show that gLRTH has the controlled type I error and power advantage over classical methods. The real data application demonstrated the high performance of gLRTH on discovering potential heterogeneous disease related loci. To further identify the casual gene, prioritization algorithms are proposed to prioritize potential disease associated SNPs, including annotation and functional enrichment analysis, and expression quantitative trait loci and enrichment analysis. To assess the predictive value of detected SNPs, we proposed a mixture model under the assumption of genetic heterogeneity and significantly improve the predictive value (20% increase in area under ROC curve).