Biological and practical implications of genome-wide association study of schizophrenia using Bayesian variable selection

Genome-wide association studies (GWAS) have identified over 100 loci associated with schizophrenia. Most of these studies test genetic variants for association one at a time. In this study, we performed GWAS of the molecular genetics of schizophrenia (MGS) dataset with 5334 subjects using multivariate Bayesian variable selection (BVS) method Posterior Inference via Model Averaging and Subset Selection (piMASS) and compared our results with the previous univariate analysis of the MGS dataset. We showed that piMASS can improve the power of detecting schizophrenia-associated SNPs, potentially leading to new discoveries from existing data without increasing the sample size. We tested SNPs in groups to allow for local additive effects and used permutation test to determine statistical significance in order to compare our results with univariate method. The previous univariate analysis of the MGS dataset revealed no genome-wide significant loci. Using the same dataset, we identified a single region that exceeded the genome-wide significance. The result was replicated using an independent Swedish Schizophrenia Case–Control Study (SSCCS) dataset. Based on the SZGR 2.0 database we found 63 SNPs from the best performing regions that are mapped to 27 genes known to be associated with schizophrenia. Overall, we demonstrated that piMASS could discover association signals that otherwise would need a much larger sample size. Our study has important implication that reanalyzing published datasets with BVS methods like piMASS might have more power to discover new risk variants for many diseases without new sample collection, ascertainment, and genotyping.