Assessment of Genomic Prediction for Sorghum Seed Parent Evaluation

Abstract

Genomic prediction has shown to be an effective tool for plant and animal breeders, allowing for prediction of phenotypes and breeding values. For crops that have vast acres planted in the United States, such as maize, soybean, and wheat, numerous studies have shown the effectiveness of genomic prediction. Consequently, both public and private breeding programs for these crops have largely adopted genomic prediction. However, relatively few studies have shown the effectiveness of genomic prediction in sorghum consequently leading to limited adoption in sorghum breeding programs. The dissertation presented herein assesses the effectiveness of genomic prediction for a sorghum breeding program, primarily focusing on sorghum B-line (female) recombinant inbred line populations. The effectiveness of genomic prediction for B-line evaluation was assessed in three distinct ways: 1) testing if genomic prediction can be used to screen sorghum B-lines for phenotypes that are important to hybrid seed production; 2) testing if genomic prediction can be used to predict the performance of sorghum B-lines in hybrid test crosses in various cross validation (CV) schemes in individual environments; and 3) testing if including genotype by environment effects to predict hybrid performance across environments improves genomic prediction accuracies and effectiveness of various multilocation CV schemes. The observations show that genomic prediction can be used to effectively predict both phenotypes of sorghum B-lines and the phenotypes of B-lines in hybrid test crosses. In particular, it was observed that sorghum B-lines can be assessed prior to phenotyping for their agronomic acceptability in regard to hybrid seed production using genomic prediction. Likewise, it was observed that genomic prediction can predict sorghum B-line performance in hybrid test crosses in individual environments. Additionally, it was observed that genomic prediction models including genotype x environment effects were capable of predicting B-line performance in hybrid test crosses across environments, although some cross validation schemes were more predictive than others.