SABO-ILSTSVR: a genomic prediction method based on improved least squares twin support vector regression.

Area of Science:

• Plant breeding
• Genomics
• Machine learning in agriculture

Background:

• Genomic prediction (GP) utilizes high-density single nucleotide polymorphisms (SNPs) for predicting genomic estimated breeding values (GEBVs), accelerating crop improvement.
• Overfitting is a common challenge in GP due to a higher number of SNPs than samples, hindering accurate predictions.

Purpose of the Study:

• To develop an optimized genomic prediction model to overcome overfitting in marker-based breeding.
• To enhance the accuracy and efficiency of genomic selection in modern plant breeding programs.

Main Methods:

• The study proposes an enhanced Least Squares Twin Support Vector Regression (LSTSVR) model, termed ILSTSVR, by integrating a Lasso regularization term.
• A novel subtraction average based optimizer (SABO) was developed to automatically tune the parameters of the ILSTSVR model, resulting in the SABO-ILSTSVR model.
• The performance of the SABO-ILSTSVR model was evaluated using four diverse crop datasets.

Main Results:

• The SABO-ILSTSVR model demonstrated superior or equivalent performance compared to existing widely-used genomic prediction methods across all tested datasets.
• The optimization approach effectively mitigated overfitting issues inherent in traditional GP models.
• The proposed method offers a robust solution for accurate GEBV prediction in plant breeding.

Conclusions:

• The SABO-ILSTSVR model represents a significant advancement in genomic prediction, offering improved accuracy and efficiency for plant breeding.
• This optimized approach provides a valuable tool for accelerating genetic gain and shortening breeding cycles.
• The study highlights the potential of integrating advanced machine learning techniques for robust genomic selection.