Prediction Accuracy Of Various Genomic Selection Models On Yield And Quality Traits In Chinese Winter Wheat

Conventional plant breeding largely relies on phenotypic selection to select superior individuals among a huge number of segregating populations.In conventional breeding,selection efficiency is low and prediction is inaccurate,especially for polygenic traits with low heritability.Advances in molecular biology and biotechnology have enabled development and application of molecular markers for genetic studies of breeding targeted traits,providing an opportunity for genotypic selection in breeding.Several molecular marker-based selection approaches such as marker-assisted recurrent selection and genomic selection have been proposed to speed up the selection procedure and improve breeding efficiency.Empirical evaluation of different selection methods is a daunting task which demands more time,more physical resources,and labour.Computer simulation could overcome some of those constraints and provide an opportunity to evaluate different selection methods under a wide range of genetic models(e.g.,additive and epistasis).The main purpose of present study is to assess the performance of different selection methods under different genetic architectures using empirical and simulated data in wheat.1.Genomic prediction for grain yield and yield-related traits in Chinese winter wheatUsing one natural wheat population consisting of 166 accessions,prediction accuracies of seven GS models were evaluated on yield and yield-related traits from various SNP quality control(QC)scenarios(missing rate and minor allele frequency),missing genotype imputation,and genome-wide association studies(GWAS)-derived markers.Results indicated that a moderate missing rate level(20% to 40%)and moderate minor allele frequency(MAF)threshold(5%)provided higher prediction accuracy.It was observed that accuracies of different traits were related to their heritability and genetic architecture,as well as the GS prediction model.Moore–Penrose generalized inverse,ridge regression,and random forest resulted in higher prediction accuracies than other GS models across traits.Imputation of missing genotypic data had marginal effect on prediction accuracy,while GWAS-derived markers improved the prediction accuracy in most cases.SNP QC on missing rate and MAF had positive impact on the predictability of GS models.2.Genomic selection for flour color-related traits in wheatWheat flour-color plays an important role in consuming quality of a wheat cultivar.Direct measurement of flour color-related traits can be expensive and time-demanding.One natural wheat population comprised of 166 accessions was used to investigate GS prediction accuracies on wheat flour color-related traits.GS model construction included GWAS-derived markers and shared QTL.The highest prediction accuracy(>0.80)was obtained when GWAS-derived markers were incorporated in GS model.Results indicated that markers selected by single-locus and multi-locus GWAS models showed comparable accuracies for five flour color-related traits.In terms of shared(or common)QTL of genetically correlated traits,including multi-trait QTL in prediction model did not improve prediction accuracy as compared with single-trait GWAS-derived QTL.3.Comparison of phenotypic recurrent,marker-assisted recurrent,and genomic selections by using the QuMARS simulation toolA QU-GENE based application module called “QuMARS” was used to compare selection responses from different selection methods including phenotypic recurrent selection,marker-assisted recurrent selection,and genomic selection(PS,MARS and GS)over 15 cycles.One additive,two QTL linkage phases,and two epistasis were developed and evaluated.Selection responses from genomic best linear unbiased prediction(GBLUP)and MARS(regression by forward selection)were consistently higher than those from PS under the additive and coupling linkage models,particularly in early cycles.In contrast,under epistatic models,PS was observed consistently superior over MARS and GS.Total genetic and additive variance was increased in some cases after selection,especially for epistatic models.GS and PS were effective recurrent selection methods for improving traits controlled by additive and epistatic quantitative trait loci(QTL).QuMARS allows breeder and researcher to compare,optimize,and integrate new technology into their conventional breeding programs.4.Factors affecting the prediction accuracy in simulated populationsA QU-GENE based application module called “QuLine” was used to simulate different genetic architectures that included three QTL numbers(1,5,and 10 per chromosome)and three heritability levels(0.3,0.6,and 0.9).Results indicated that the increase in QTL number improved prediction accuracy,especially for 5 and 10 QTL per chromosome in later cycles.Prediction accuracies from GS models were similar across QTL numbers and heritability levels.In most cases,5 cM marker interval resulted in higher prediction accuracy than 10 cM marker interval.Marker interval with a mean distance between markers of 5 cM provided larger marker number as well as higher prediction accuracy.