Genomic predictions to leverage phenotypic data across genebanks.

Area of Science:

• Agricultural Science
• Genetics
• Plant Breeding

Background:

• Genome-wide prediction is crucial for breeding and utilizing genebank resources.
• Plant genetic resources in genebanks often have information gaps, limiting their use in prebreeding.
• Across-genebank prediction can unlock the potential of these valuable collections.

Purpose of the Study:

• To evaluate the effectiveness of across-genebank prediction for barley (Hordeum vulgare L.) germplasm.
• To compare different prediction scenarios, including within-genebank and across-genebank approaches.
• To assess the impact of training set size and composition on prediction accuracy.

Main Methods:

• Utilized historical data on flowering/heading date, plant height, and thousand kernel weight for 9,344 barley accessions from IPK and 1,089 from ICARDA.
• Compared three prediction scenarios: benchmark (within ICARDA), across-genebank (IPK training, ICARDA test), and integrated (IPK + 30% ICARDA training, ICARDA test).
• Analyzed prediction abilities across different scenarios to determine the most effective approach.

Main Results:

• Within-genebank predictions showed low to moderate accuracy, likely due to limited training data.
• Across-genebank predictions using integrated training sets (IPK + 30% ICARDA) boosted prediction abilities up to ninefold.
• Genotype × environment interactions (GEIs) posed a challenge, but were counterbalanced by larger, connected training sets.

Conclusions:

• Across-genebank prediction, particularly with integrated training sets, significantly improves the predictive characterization of plant genetic resources.
• This approach can enhance the curation of global genebank collections and advance the development of biodigital resource centers.
• Augmenting training data across genebanks overcomes limitations of small, isolated datasets and potentially mitigates GEI effects.