Large-scale genomic prediction using singular value decomposition of the genotype matrix

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Genomic prediction models face computational challenges with increasing numbers of loci and individuals.
• Calculating the inverse genomic relationship matrix (GRM) for large datasets is computationally intensive.
• There is a significant need for dimensionality-reduction techniques to analyze massive genomic data.

Purpose of the Study:

• To develop and evaluate reduced-dimension singular value decomposition (SVD) based models for efficient genomic prediction.
• To address the computational demands of analyzing large-scale genomic datasets.

Main Methods:

• Developed reduced-dimension SVD models for genomic prediction.
• Utilized chromosome-wise or segment-specific SVD to generate principal components (PCs).
• Applied PC ridge regression (PCRR) or genomic animal models with inverse GRM from PCs (PCIG) for prediction.

Main Results:

• PCRR and PCIG models using chromosome-wise SVD accurately predicted breeding values, yielding results virtually identical to full-dimension models (r=1.000).
• These SVD-based models demonstrated robustness to core sample size, performing well even with as few as 500 individuals.
• The method was successfully validated on a large, multi-breed dataset.

Conclusions:

• Singular value decomposition (SVD) effectively reduces dimensionality in large genomic datasets.
• SVD enables computationally efficient genomic prediction with dense genomic data and numerous individuals.
• The developed method produces genomic estimated breeding values virtually identical to those from full-dimension models.