Efficient Genomic Control for Mixed Model Associationsin Large-scale Population

Abstract

Among linear mixed model-based association methods, GRAMMAR has the lowest computing complexity for association tests, but it produces a high false-negative rate due to the deflation of test statistics for complex population structure. Here, we present an optimized GRAMMAR method by efficient genomic control, Optim-GRAMMAR, that estimates the phenotype residuals by regulating downward genomic heritability in the genomic best linear unbiased prediction. Even though using the fewer sampling markers to evaluate genomic relationship matrices and genomic controls, Optim-GRAMMAR retains a similar statistical power to the exact mixed model association analysis, which infers an extremely efficient approach to handle large-scale data. Moreover, joint association analysis significantly improved statistical power over existing methods.