A joint learning approach for genomic prediction in polyploid grasses

Area of Science:

• Plant genetics and breeding
• Bioinformatics and computational biology
• Agricultural science

Background:

• Poaceae family plants, including sugarcane and forage grasses, are vital economically but possess complex polyploid genomes.
• Limited genetic resources and genomic complexity in these species hinder advanced breeding strategies like marker-assisted selection.
• Genomic selection offers potential for genetic gain, but existing models are inadequate for polyploid species.

Purpose of the Study:

• To develop an advanced machine learning method for accurate trait prediction in polyploid Poaceae species.
• To enhance genomic selection models by integrating a joint learning approach for complex trait prediction.
• To improve breeding efficiency and accelerate genetic gains in economically important polyploid crops.

Main Methods:

• Genotyping of biparental populations of sugarcane (Saccharum spp.), Urochloa decumbens, and Megathyrsus maximus.
• Measurement of various quantitative traits across different cross-validation scenarios.
• Development and application of a machine learning system combining classification and regression for trait prediction using high-quality markers.

Main Results:

• The developed joint learning approach demonstrated significant predictive capability for complex traits.
• The proposed machine learning strategy achieved prediction accuracy improvements exceeding 50% compared to traditional genomic prediction methods.
• The system showed promising results in predicting traits from genotypic data in polyploid species.

Conclusions:

• The novel machine learning methodology offers a powerful tool for genomic selection in polyploid crops.
• Implementation in breeding programs can substantially reduce breeding cycles and enhance genetic gains.
• This approach addresses the challenges posed by genomic complexity in important agricultural species.