Training Population Optimization for Genomic Selection in Miscanthus.

Area of Science:

• Plant breeding and genetics
• Bioenergy crop development
• Genomic selection applications

Background:

• Miscanthus × giganteus (M×g) is a key biofuel feedstock, but limited genetic diversity hinders breeding.
• Genomic selection (GS) offers a path to improve biofuel traits in M×g.
• Utilizing parental species Miscanthus sinensis (Msi) and Miscanthus sacchariflorus (Msa) germplasm for GS training is a potential strategy.

Purpose of the Study:

• To evaluate the effectiveness of Msi and Msa diversity panels as training sets for GS in M×g.
• To identify optimal training set composition for accurate genomic prediction of M×g breeding values.
• To understand the impact of genetic architecture on GS prediction accuracy in interspecific populations.

Main Methods:

• Assessed GS prediction accuracies within Msi and Msa panels, considering subpopulation structure.
• Trained GS models using Msi and Msa subsets to predict breeding values in an M×g F2 panel.
• Evaluated prediction accuracies in simulated M×g F2 panels with varying parental genetic contributions.

Main Results:

• Subpopulation structure significantly impacted GS accuracies within Msi and Msa panels.
• GS models trained on Msi and Msa showed low and negative prediction accuracies for M×g.
• Genetic architectures with shared causal mutations across parental species resulted in the highest prediction accuracies.

Conclusions:

• Training sets for M×g genomic selection should ideally mirror the genetic architecture of the target population.
• Maximizing genetic relatedness between training and validation sets is crucial for accurate genomic prediction.
• Future breeding efforts should focus on curating diverse training sets that capture causal mutations relevant to M×g.