Fig. 1

The procedure of E-GWAS. No single method can be entirely optimal for all traits due to the diversity of the genetic architectures of complex traits. For many traits, the putative SNPs detected by different methods do not completely overlap. Our proposed E-GWAS strategy integrates different GWAS methods to adapt different traits through a three-step procedure. First, we identify the overlapping SNPs between lists of SNPs for each pair of N GWAS methods and combine them to obtain a preliminary combined list with m SNPs. Because some putative SNPs identified by the different methods might be close to each other, a within-bin merged method can be used to expand the size of the intersection windows (i.e., 10 and 50 kb) (a). Then, the m SNPs are simultaneously integrated into a mixed-effect linear model as fixed effects, and their p-values are calculated (b). Among the m putative SNPs, SNPs with a Pearson correlation coefficient of their genotypes greater than 0.7 are clustered together and the j remaining SNPs are retained, those with the lowest p value in each of the j clusters (c). The j SNPs are again fitted in a mixed-effect linear model (d). The permutation test method is used to define the threshold for the j p-values corresponding to the fitted SNPs. The phenotype (y) is shuffled n times and the p-values of the j SNPs are re-calculated. Then, the 0.01-quantile of the n minima of the j p-values in n times is defined as the threshold (e). Finally, i SNPs with p-values below the threshold are retained