Genome-wide association studies (GWASs) aim to detect genetic risk factors for complex human diseases by identifying disease-associated single-nucleotide polymorphisms (SNPs). SNP-wise approach, the standard method for analyzing GWAS, tests each SNP individually. Then the P-values are adjusted for multiple testing. Multiple testing adjustment (purely based on p-values) is over-conservative and causes lack of power in many GWASs, due to insufficiently modelling the relationship among SNPs. To address this problem, we propose a novel method, which borrows information across SNPs by grouping SNPs into three clusters. We pre-specify the patterns of clusters by minor allele frequencies of SNPs between cases and controls, and enforce the patterns with prior distributions. Therefore, compared with the traditional approach, it better controls false discovery rate (FDR) and shows higher sensitivity, which is confirmed by our simulation studies. We re-analyzed real data studies on identifying SNPs associated with severe bortezomib-induced peripheral neuropathy (BiPN) in patients with multiple myeloma. The original analysis in the literature failed to identify SNPs after FDR adjustment. Our proposed method not only detected the reported SNPs after FDR adjustment but also discovered a novel SNP rs4351714 that has been reported to be related to multiple myeloma in another study.