Factors Affecting Genomic Prediction Accuracy In Soybean Using Genotyping-By-Sequencing.

Advancements in high-density genotyping combined with powerful statistical methods created  what is known as “genomic selection”, a breeding methodology with great potential to expedite genetic gain. The potential of genomic selection for soybean, the second most valuable crop in the United States, has not been studied to our knowledge. The objective of this research was to evaluate the accuracy of genomic predictions in a soybean breeding program. Factors such as choice of markers based on percentage missing data and minor allele frequency, marker imputation, training population size, and statistical model were studied. University of Nebraska breeding lines were scored for 63,228 SNPs (minor allele frequency > 0.05) uniformly covering the soybean genome using genotyping-by-sequencing methods.  Fifty percent of the SNPs displayed less than 30% missing data. Using cross validation, we show that prediction accuracies for yield, maturity date, and height can be quite high for a range of marker subsets. Little difference was observed between statistical models. A random forest imputation method provided modest increases in accuracy. Results suggest that when using GBS data for genomic predictions in soybean, accuracy is maximized by including all SNPs despite missing data frequency, and performing imputation on missing marker data points.