Genomic Selection By Pre-selection Of Markers

In modern animal breeding, larger genetic progress can be obtained by genomic selectionusing high-density markers covering genome. However, high-density marker panels will bringsome drawbacks:(1) Some SNPs with non-effect may be estimated with effects, which willreduce the accuracy of genomic estimated breeding values (GEBVs);(2) There may be strongcollinearity between markers, however, epistatic effects are not calculated in genomicselection. Thus, higher density markers may bring more errors;(3) More time needs to bewasted to calculate effects of high-density markers, which brings a large challenge inalgorithms;(4) It’s not cost effective to use high-density marker panels to estimate GEBVs,which is not helpful to apply genomic selection in small domestic animals. Low-densitymarker panels can make up above drawbacks in algorithms and economy. Thus, in thisinvestigation, we will try to select some markers (SNPs) associated with traits from genometo estimate GEBVs.In the first experiment, we simulated natural population by computer as a trainingpopulation. SMA (single marker analysis) or SME (selecting markers according to effects)was used to select SNPs. emBayesB or GBLUP algorithm was used to estimate GEBVs. Inthis simulation, we set different parameters, e.g., heritability and number of QTL affectingtraits, and simulated6generations as validation population. The methods of selecting SNPsand results are:(1) When using all SNPs, higher heritability brings higher accuracy whatever emBayesB orGBLUP was used to estimate GEBVs, which showed that heritability was one of the factorsaffecting accuracy of GEBVs. The accuracy of GEBVs by emBayesB reduced when thenumber of QTL became larger. The reason was that mean genetic variance of each QTLbecame smaller with more QTL existing. However, this condition didn’t happen in GBLUPalgorithm. The reason was GBLUP was more suitable to the situation that more QTL exist ingenome.(2) SMA used all SNPs to associate with phenotypes, and made standards of selectingSNPs according the P values (F values) after significance test. SME used emBayesB tocalculate the effects of all markers and selected markers according to their effects (variances).The results showed that SME was suitable to combine with emBayesB while SMA wassuitable to combine with GBLUP to make low-density panels for genomic selection. IfGBLUP was used to estimate GEBVs, the results after selecting markers were better thanbefore whatever we used SMA or SME to select SNPs. However, SMA was more suitable to be used when the number of remaining markers was large. When fewer markers wereremained, the results of SME were better than that of SMA.(3) Whatever we used SME or SMA to select markers with purposes, the results were muchbetter than random selecting markers, which showed that these methods had somesignificance for selecting markers with effects.The second experiment tried to compare the results of R-SMA (restricting number ofmarkers in some region) and simple SMA. The results showed that when fewer SNPs wereremained, the accuracy of GEBVs by R-SMA was higher than that obtained by SMA, whichfurther showed that SMA might ignore markers with small effects when finding markersassociating with traits. R-SMA can provide references for breeders who use lower-densitypanels to develop genomic selection.All results show that genomic selection by pre-selecting SNPs may improve accuracy ofGEBVs. Considering no increasing marker density and number of training population andbasically ignoring cost of computing and saving cost of breeding in continuous generationsafter making low-density panels, using low-density panels by pre-selecting markers can offerreference for molecular breeding and genomic selection in small domestic animals.