Haplotypes and Machine Learning Improve Genomic Prediction in Brassica napus

Area of Science:

• Plant breeding
• Genomics
• Bioinformatics

Background:

• Genomic selection (GS) aids crop and livestock improvement by predicting complex traits.
• Genomic prediction (GP) models face challenges in accounting for genomic interactions like epistasis.
• Haplotypes and Machine Learning (ML) are methods to address epistasis and non-linear relationships in GP.

Purpose of the Study:

• To compare the effectiveness of linear and ML-based GP models using single nucleotide polymorphisms (SNPs) and haplotypes in Brassica napus.
• To evaluate the impact of incorporating haplotypes and ML on trait prediction accuracy.
• To assess the potential of these methods for improving GS in breeding programs.

Main Methods:

• Utilized a publicly available dataset of 991 Brassica napus individuals with 4,286,896 SNPs.
• Developed and compared linear and ML-GP models using both SNP and haplotype data.
• Assessed prediction accuracy for flowering time, oil content, and oleic acid content.

Main Results:

• ML-based GP models demonstrated improved trait prediction accuracy compared to linear models for all tested traits.
• Both SNP- and haplotype-based GP models benefited from ML approaches.
• Haplotypes, while not significantly boosting accuracy over SNPs, captured novel genetic variation and offered broader diversity.

Conclusions:

• ML significantly enhances GP model performance in Brassica napus for key agronomic traits.
• Haplotypes provide a qualitative advantage for long-term breeding by increasing genetic diversity.
• Combining ML and haplotypes presents a promising strategy for advancing genomic selection in Brassica napus.