Genome-wide prediction of traits with different genetic architecture through efficient variable selection

Area of Science:

• Genomics
• Quantitative Genetics
• Plant Breeding

Background:

• Genome-based prediction accuracy relies on appropriate statistical models, with variable selection methods (e.g., LASSO) potentially outperforming methods like RR-BLUP for traits with few quantitative trait loci (QTL).
• Investigating the conditions for successful variable selection is crucial for advancing genomic prediction in diverse populations.

Purpose of the Study:

• To evaluate the assumptions and effectiveness of variable selection methods in genome-based prediction.
• To compare variable selection with methods like RR-BLUP across different species and datasets.

Main Methods:

• Combined computer simulations with large-scale experimental data from rice, wheat, and Arabidopsis thaliana.
• Analyzed the impact of sample size, heritability, and linkage disequilibrium (LD) on prediction accuracy.

Main Results:

• Variable selection is effective when the number of phenotyped individuals significantly exceeds the number of causal mutations.
• Required sample size for variable selection increases with decreasing heritability and increasing LD.
• In plant breeding populations, superiority of variable selection over RR-BLUP is unlikely due to long-range LD, medium heritabilities, and small sample sizes.

Conclusions:

• The effectiveness of variable selection methods is highly dependent on specific genetic and population parameters.
• Findings impact the choice of statistical methods for genetic architecture and prediction accuracy in plant breeding and whole-genome sequence data analysis.