Efficient Weighting Methods For Genomic BLUP Adaption To The Genetic Architectures Of Quantitative Traits

Genomic best linear unbiased prediction(GBLUP)assumes equal variance for all marker effects,this is suitable to traits conform to the infinitesimal model.For traits controlled by major genes,Bayesian methods with shrinkage priors or genome-wide association study(GWAS)methods can be used to identify causal variants effectively.The information from Bayesian/GWAS methods can be used to construct the weighted genomic relationship matrix(G).Currently,the research on this issue is insufficient in some aspects:(a)the influence of marker information sources on weighting methods was not adequately investigated;(b)systematic and theoretical construction methods have not been reported.In this study,we developed methods to optimize the efficiency of weighted GBLUP for traits with different genetic architectures.Main contributions of current study are summarized as follow:(1)Two types of methods(marker effect with local-shrinkage or normal prior)were used to get the test statistics and the effect estimate for each marker.Second,three weighted G matrices were constructed based on the marker information from the first step:(?)the genomic-feature weighted G(GFWG),(?)the estimated marker-variance weighted G(EVWG),and(?)the absolute value of estimated marker-effect weighted G(AEWG).Six options of weighted GBLUP(local-shrinkage/normal prior GF/EV/AE-WGBLUP)were produced from the two steps.(2)The heritability(h2)and the number of quantitative trait loci(QTL)are considered as two factors that influencing the genetic architecture of quantitative traits.And a series of simulated traits were produced in our simulation study.Compare to GBLUP,the gain of weighting methods in accuracy was trait dependent,ranging from 14.7% to marginal for simulated traits.Local-shrinkage prior EVWGBLUP is superior for traits mainly controlled by loci with large effect.Normal prior AEWGBLUP performs well for traits mainly controlled by loci with moderate effect.For traits mainly controlled by loci with small effect and also influenced by some loci with large effect,GFWGBLUP based on powerful GWAS methods has advantages.For traits controlled by loci with small effect,our weighting methods are similar to that of GBLUP.(3)The weighting methods we proposed were applied in 25 real traits from dairy cattle,pig and loblolly pine.The results demonstrated that these weighting methods offer flexibility for optimizing the weighted GBLUP for traits with a broad realm of genetic architectures.Compare to GBLUP,the gain of weighting methods in accuracy ranged from 44% to marginal for real traits.Compared with the published results,our methods have advantages in most cases,which verifies the effectiveness of our methods.In conclusion,different weighted GBLUP were constructed in our study,which allows the marker information obtained from GWAS/GP methods to be used in different ways.Corresponding to the Bayesian alphabet,the combinations of different weighted GBLUP and GWAS/GP methods can adapt to various genetic architectures.Meanwhile,our weighting methods have the advantage of easily integrating into the existing genetic evaluation infrastructure that uses pedigree to derive a relationship matrix.