Related Experiment Video
Updated: Jun 22, 2026
07:09
A Rapid and Efficient Method for Assessing Pathogenicity of Ustilago maydis on Maize and Teosinte Lines
Published on: January 3, 2014
Maximizing efficiency of genomic selection in CIMMYT's tropical maize breeding program
Sikiru Adeniyi Atanda1,2,3, Michael Olsen4, Juan Burgueño2
1West Africa Center for Crop Improvement (WACCI), University of Ghana, Accra, Ghana.
Summary
Optimizing training sets with historical data significantly boosts genomic selection accuracy in maize breeding. Selecting informative individuals improves prediction, shortening breeding cycles and reducing resource needs.
Area of Science:
- Plant breeding
- Quantitative genetics
- Genomic selection
Background:
- Current genomic selection (GS) strategies in maize breeding rely on large full-sib populations, limiting breeding cycle times.
- The "test-half-predict-half" approach, while effective, has constraints in population size and cycle time reduction.
- Efficiently utilizing historical breeding data is crucial for enhancing GS accuracy and program efficiency.
Purpose of the Study:
- To identify optimal experimental and training set designs for maximizing prediction accuracy in CIMMYT's maize breeding programs.
- To evaluate different training set (TS) design strategies for genomic prediction (GP) using historical phenotypic data.
- To improve the efficiency of GS by optimizing the use of phenotyping resources.
Main Methods:
- Evaluation of training set design strategies using datasets of 849 (DS1) and 1389 (DS2) doubled haploid (DH)-lines evaluated as testcrosses.
- Utilized algorithms to select individuals for training sets that maximize relatedness between training and prediction sets.
- Compared prediction accuracy across different training set compositions and phenotyping strategies.
Main Results:
- Using multiple bi-parental populations as training sets, selected for maximum relatedness, significantly improves prediction accuracy.
- Spreading phenotyping expenditure across connected bi-parental populations enhances prediction accuracy compared to within-population prediction, especially with small training sets.
- Optimizing line selection for phenotyping versus genotyping improves prediction accuracy in sparse testing and "test-half-predict-half" scenarios.
Conclusions:
- Historical breeding data can be efficiently leveraged to enhance genomic selection accuracy through optimized training set design.
- Strategic selection of informative individuals within and across populations is key to maximizing predictive performance.
- Optimized experimental designs reduce resource requirements and accelerate breeding cycle times in large-scale maize improvement programs.
