Machine learning for genomic and pedigree prediction in sugarcane

Area of Science:

• Plant Breeding
• Genomics
• Machine Learning
• Quantitative Genetics

Background:

• Sugarcane breeding programs face challenges due to the highly heterozygous polyploid genomes of Saccharum spp.
• Accurate genome prediction (GP) is crucial for efficient crop improvement, especially in polyploid species.
• Non-additive genetic effects are significant in polyploid crops and need to be considered in GP models.

Purpose of the Study:

• To incorporate non-additive genetic effects and pedigree information into GP models for sugarcane.
• To evaluate the effectiveness of machine learning methods in enhancing prediction accuracy for sugarcane breeding lines.
• To compare the performance of machine learning approaches against traditional methods like Best Linear Unbiased Prediction (BLUP).

Main Methods:

• Genotyping of 297 sugarcane clones using 33,149 single-nucleotide polymorphisms (SNPs).
• Implementation of BLUP, random forests, and a novel simulation annealing ensemble (SAE) method.
• Validation using repeated 10-fold cross-validation and leave-one-family-out cross-validation strategies.

Main Results:

• Machine learning methods, particularly SAE, showed higher prediction accuracy than BLUP across different cross-validation schemes.
• Repeated 10-fold cross-validation indicated machine learning methods consistently outperformed BLUP.
• Leave-one-family-out cross-validation highlighted SAE as the most accurate prediction method.

Conclusions:

• Genome prediction is effective for Japanese sugarcane breeding programs.
• Machine learning methods offer significant potential for improving prediction accuracy in sugarcane by modeling complex genetic interactions.
• The study validates the utility of advanced computational approaches for accelerating sugarcane genetic improvement.