Hierarchical Solution Method Of Genome-Wide Association Study

The genetic basis of quantitative trait is the genes controlling the trait.The earlier polygenic hypothesis suggested that quantitative traits are the result of the action of many independent genes,each of which has little effect on the traits,called polygenes.In practice,by constructing genetic separation groups and phenotypic separation analysis,it is proved that many genes controlling quantitative traits have different effects,and the main effect genes are widespread.A large number of the experimental results about GWAS show that the genetic variation of some quantitative traits mainly depends on the major genes,and some of them depend on the polygenes,but most of the case is determined jointly by the major genes and polygenes.Thus,it is the theoretical foundation for precisely mapping QTNs and accurately estimate genomic breeding values that properly dissecting the genomic variation of quantitative trait.This study partitions the genomic variation into major gene effects and polygenic effects,constructing a unified genetic model to analyze the genomic variation of quantitative trait.We propose a hierarchical algorithm here,called GEBV-one,to solve the genetic model,that is firstly estimating the breeding values,namely the sum of major gene effects and polygenic effects with GBLUP and then with the estimated breeding values as the dependent variables,conducting regression analysis SNP by SNP using PLINK software.The significance of each SNP effect obtained by regression analysis can be statistically inferred by the test statistics constructed by SNP effects and random errors.Using the computer simulations and a series of published datasets about GWAS,we can systematically demonstrate the efficiency,reliability and adaptability of the new method.In addition,we also compare the new method with the several popular simplified solution strategies of mixed models for GWAS,such as PLINK,EMMAX,GRAMMAR and FaST-LMM algorithms.Comparison results show that:(1)Without taking into account the influence of other SNPs on the target SNP,the PLINK algorithm can not accurately estimate the effects of SNPs and exhibits a high false positive rate.Since the genomic estimated breeding value(GEBV)has included the variance of the main effect gene,the EMMAX method increases the estimation of the heritability of the trait and reduces the detection efficiency of the main effect gene.The statistical detection efficiency of the GRAMMAR algorithm is completely dependent on the estimating accuracy of GBLUP for GEBV.The lower the accuracy,the higher the detection efficiency.The FaST-LMM method can detect all SNPs by only one spectral decomposition,which greatly reduces the demand for computer memory while reducing computation time.(2)Within the framework of linear mixed model,the GEBV-one algorithm can powerfully identify the major variants along with estimating the heritability in high goodness of fit to phenotypes.For the real dataset of growth traits in Japanese flounder,the theoretical method can be utilized to conduct GWAS and genomic selection,reveal the genetic patterns of these traits and assist the selective breeding of the growth traits based on the estimated breeding values.