Support vector machine regression for the prediction of maize hybrid performance

Area of Science:

• Agricultural Science
• Genetics
• Machine Learning

Background:

• Accurate prediction of hybrid performance accelerates genetic gain and reduces breeding costs.
• Existing methods like best linear unbiased prediction (BLUP) are effective but may lack flexibility.

Purpose of the Study:

• To introduce and evaluate epsilon-insensitive support vector machine regression (epsilon-SVR) for predicting phenotypical performance of untested single-cross hybrids.
• To compare the predictive accuracy of epsilon-SVR with BLUP using real-world data.

Main Methods:

• Utilized epsilon-SVR with kernel functions based on dominant and co-dominant genetic similarity measures.
• Integrated diverse marker types into a single regression model via kernel operations.
• Applied the method to grain maize breeding data from RAGT R2n.

Main Results:

• Epsilon-SVR demonstrated comparable prediction accuracies to BLUP across various marker types and traits.
• The epsilon-SVR framework showed greater flexibility in integrating different predictor variables.

Conclusions:

• Epsilon-SVR is a viable and flexible alternative for predicting hybrid crop performance.
• This method can enhance genetic progress in breeding programs by improving prediction accuracy and variable integration.