BLINK AD: A Multi-Locus Model for Detecting Additive and Dominance Effects in Genome-Wide Association Studies

During the past decade, genome wide association study(GWAS) has been widely used for mapping genes in nature populations. Linear mixed model(LMM) is widely used to eliminate the false positive caused by population structure. However big data leads to extreme computation challenge, especially when non-additive effects were considered in the model. Currently, only a few methods can perform genome wide additive and dominance(AD) effects association analysis. Our group has newly developed an ultra-fast and powerful multi-locus additive model named BLINK. Here we implemented dominance effect into BLINK model. Simulation study showed that our model has higher power compared with PLINK AD model and Bolt-LMM under different scenarios. In a Finland human population, we analyzed 5 traits with significant dominance effects which were calculated by GREMLd model implemented in GCTA. BLINK AD model could find much more association loci, indicating a potential complex regulation mechanism for those traits. Our results also showed that most of the significant SNPs have complete dominance or over-dominance effects. Furthermore, we performed the prediction analysis using Bayesian LASSO and gBLUP. Results showed that prediction accuracies of Bayesian LASSO using GWAS signifcant loci are higher than gBLUP using whole genome SNPs. We also observed similar results from a rice hybrid population. We anticipate our model would help elucidate the genetic mechanism of complex traits and be useful for genomic selection in plant and animal breeding.