A Quick Approximation Method For The Association Analysis Of The Whole Genome Generalized Mixed Model

In genome-wide association study(GWAS),the inflation of test statistics caused by population stratification,family structure and cryptic relation increases false positive rate for detecting quantitative trait nucleotides(QTNs).Simple principal component analysis(PCA)and comprehensive linear mixed model(LMM)have been applied into genomic control(GC)for those confounders.With spectral decomposition of released relationship matrix derived from whole genomic markers,the LMM of polygenes is shown to be equivalent to that of the weighted principal components(PCs)by corresponding squared roots of eigenvalues.This suggests all PCs have ability to evaluate other confounders,in addition to population stratification.Based on this theory,we select a number of the top PCs to correct confounders according to genome-wide chi-squared mean(close to 1)or a satisfied Q-Q plot.The difference between disease traits and conventional traits is that disease traits need to first select the principal components through the generalized linear model(GLM),and then use the selected principal components as the covariates of the generalized linear model to return,while the conventional quantitative traits can be directly perform PCs phenotypic correction.The method by using the principal component effectively rectified by population stratification,kinship hybrid effect,caused by the correlation analysis of generalized linear mixed models(GLMM)into the generalized linear regression model(GLM),and then use PLINK,efficiently genome-wide association analysis software,to locate gene for the binary traits and achieve rapid approximate mixed model the effect of correlation analysis.For demonstrating the proposed method,the human data set of 4,000 individuals with 500,000 markers was used for the simulation correlation analysis.The simulation results show the superiority of the new optimization method,but there is still a slight gap with the fast-lmm method.Furthermore,further correlation analysis was performed on 5 human diseases,and the results obtained were similar to the simulation results.It can correct confounding factors and improve the detection ability of QTNS.Only this verifies the effectiveness of the method.Although the effect of the new method cannot be exactly the same as that of the mixed model,the advantage of the new optimization method is that its principle is more easy to understand.Meanwhile,this study also selects partial genomic markers instead of full markers to calculate the relationship matrix.The new method can still get good results.This means that the new method can optimize the calculation process,reduce time,and facilitate the use of workers.