Sparse testing designs for optimizing resource allocation in multi-environment cassava breeding trials

Area of Science:

• Agricultural Science
• Plant Breeding
• Genetics

Background:

• Developing improved crop cultivars necessitates multi-environment trials (METs) to assess genotype performance across diverse conditions.
• High phenotyping costs in METs restrict the evaluation of numerous genotypes in all target environments.

Purpose of the Study:

• To investigate the effectiveness of sparse testing strategies in cassava breeding programs for reducing phenotyping expenses in METs.
• To evaluate prediction models incorporating genomic data and genotype-by-environment interaction (G × E) for optimizing sparse testing designs.

Main Methods:

• Utilized a population of 435 cassava genotypes evaluated across five Nigerian environments for dry matter and fresh root yield.
• Developed sparse testing designs based on non-overlapping (NOL) and completely overlapping (OL) genotype allocations.
• Assessed three prediction models: one phenotype-only and two incorporating genomic data, with and without G × E modeling.

Main Results:

• All models demonstrated higher predictive ability and lower mean square error (MSE) with larger training datasets.
• Modeling G × E significantly improved predictive ability and reduced MSE for given training set sizes.
• Increased OL genotypes led to decreased predictive ability and increased MSE, indicating a need for minimal OL genotypes in training sets.

Conclusions:

• Sparse testing, particularly when incorporating G × E, offers a viable approach to reduce phenotyping costs in cassava METs.
• Optimizing the size and distribution of training populations in sparse testing can enhance predictive ability and cost-efficiency.
• Integrating crop growth models (CGMs) with genomic prediction presents future potential for further improving predictive accuracy.