Approximate Genome-Based Kernel Models for Large Data Sets Including Main Effects and Interactions.

Area of Science:

• Genomics
• Quantitative Genetics
• Bioinformatics

Background:

• Genomic prediction (GP) and selection (GS) leverage molecular markers and sequencing for breeding.
• Large-scale genomic datasets (millions of observations) present computational challenges due to large kernel matrices.
• Genomic x environment interaction and multi-trait models exacerbate these computational difficulties.

Purpose of the Study:

• To develop and evaluate an approximate kernel method for reducing computational burden in GP.
• To decrease the exponential increase in computing time associated with large genomic relationship matrices.
• To assess the trade-off between computational efficiency and prediction accuracy.

Main Methods:

• Proposed an approximate kernel construction by selecting a subset of lines (m < n).
• Developed full genomic (FGSE, FGGE) and approximate (APSE, APGE) methods for single environment and genotype x environment models.
• Applied methods to wheat datasets using Genomic Best Linear Unbiased Predictor (GBLUP) and eigenvalue decomposition.

Main Results:

• Approximate methods (APSE, APGE) demonstrated competitive prediction performance compared to full methods (FGSE, FGGE).
• Significant reductions in computing time were achieved using the approximate kernel approach.
• Prediction accuracy of approximate methods is influenced by eigenvalue decay and the size of the selected subset (m).

Conclusions:

• The approximate kernel method offers a computationally efficient alternative for genomic prediction with large datasets.
• This approach maintains high prediction accuracy, making it suitable for practical breeding applications.
• The selection strategy for the subset of lines (m) is crucial for balancing accuracy and computational gains.