MAK: a machine learning framework improved genomic prediction via multi-target ensemble regressor chains and...

Area of Science:

• Quantitative genetics
• Machine learning
• Genomic prediction

Background:

• Accurate prediction of genomic estimated breeding values (GEBVs) is crucial for genetic improvement in breeding programs.
• Multi-trait models improve prediction accuracy by leveraging genotypic and phenotypic information from correlated traits.
• A key challenge is the frequent absence of phenotypic data for individuals requiring evaluation, especially in early life stages.

Purpose of the Study:

• To develop a novel machine learning framework (MAK) for enhanced GEBV prediction using only genotypic information.
• To address the issue of missing phenotypic data by constructing multi-target ensemble regression chains.
• To automatically select informative assistant traits to improve prediction accuracy of the target trait.

Main Methods:

• Proposed a machine learning framework, MAK, for GEBV prediction.
• Constructed multi-target ensemble regression chains to model relationships between traits.
• Implemented an automatic assistant trait selection mechanism within the framework.
• Evaluated MAK on four real animal and plant datasets.

Main Results:

• MAK demonstrated significantly more robust prediction ability compared to genomic best linear unbiased prediction (GBLUP), BayesB, BayesRR, and multi-trait Bayesian methods.
• The framework successfully predicted GEBVs using only genotypic data.
• MAK exhibited approximately 100 times greater computational efficiency than BayesB and BayesRR.

Conclusions:

• MAK provides a powerful and efficient approach for GEBV prediction, particularly when phenotypic data is scarce.
• The framework's ability to leverage genotypic information and select assistant traits offers a significant advancement in breeding applications.
• MAK represents a computationally efficient and robust alternative to existing prediction methodologies.