A fast genomic selection approach for large genomic data.

Area of Science:

• Genomics
• Quantitative Genetics
• Bioinformatics

Background:

• Genomic best linear unbiased prediction (GBLUP) is standard for whole-genome prediction in breeding.
• Restricted maximum likelihood (REML) used in GBLUP has high computational complexity (O(n^3)), limiting its scalability.
• Efficient methods are needed for large-scale genomic data analysis.

Purpose of the Study:

• To introduce a novel computational method, HE-BLP, for genomic selection.
• To improve computational efficiency for large genomic datasets.
• To evaluate the performance of HE-BLP against traditional GBLUP.

Main Methods:

• Developed HE-BLP by combining identical-by-state (IBS)-based Haseman-Elston (HE) regression and best linear prediction (BLP).
• Estimated total genetic variance using HE linear regression with O(n^2) complexity.
• Conducted Monte Carlo simulations and evaluated HE-BLP on a maize kernel row number (KRN) trait dataset.

Main Results:

• HE-BLP estimates total genetic variance efficiently, suitable for large datasets where environmental effects are not simultaneously estimated.
• Heritability estimates from HE regression were comparable to REML.
• HE-BLP demonstrated prediction accuracy similar to GBLUP in simulations and maize KRN trait analysis.

Conclusions:

• HE-BLP offers a computationally efficient alternative for genomic selection in large-scale datasets.
• This method is particularly useful when environmental effects can be ignored.
• HE-BLP represents a promising future direction for genomic selection in specific applications.