Dissimilarity based Partial Least Squares (DPLS) for genomic prediction from SNPs

Area of Science:

• Plant breeding and genetics
• Genomics
• Statistical modeling

Background:

• Genomic prediction (GP) enables selection of plants and animals for desirable traits without extensive field trials.
• Dissimilarity-based Partial Least Squares (DPLS) is proposed as a novel method for GP.
• The study focuses on predicting Bacterial wilt (BW) in tomatoes using SNPs.

Purpose of the Study:

• To evaluate the performance of DPLS for genomic prediction of Bacterial wilt in tomatoes.
• To compare DPLS with existing methods like Genomic Best-Linear Unbiased Prediction (GBLUP) and single-SNP regression.
• To explore the utility of DPLS in visualizing genotype-phenotype relationships for breeding selection.

Main Methods:

• Eight genomic distance measures were employed to quantify relationships between tomato accessions using SNPs.
• DPLS models were utilized with these distance measures to predict Bacterial wilt.
• Performance was assessed by comparing DPLS against GBLUP and single-SNP regression.

Main Results:

• DPLS demonstrated robustness across different genomic distance measures, yielding similar prediction performances.
• DPLS significantly outperformed single-SNP regression, indicating Bacterial wilt is a complex trait influenced by multiple loci.
• DPLS achieved prediction quality comparable to GBLUP, with the added benefit of 2-D visualization (score-plots) for candidate selection.

Conclusions:

• DPLS is a highly suitable method for genomic prediction, offering performance on par with GBLUP and superior to single-SNP regression.
• The DPLS method is versatile, compatible with various genomic dissimilarity measures and genotype representations.
• DPLS facilitates insightful data visualization, aiding breeders in selecting optimal candidates for future breeding programs.