A Benchmarking Between Deep Learning, Support Vector Machine and Bayesian Threshold Best Linear Unbiased Prediction...

Area of Science:

• Agricultural Science
• Genetics
• Plant Breeding

Background:

• Genomic selection is transforming plant breeding by enabling marker-assisted selection.
• Accurate prediction of ordinal phenotypes is crucial for selecting superior genotypes.
• Existing statistical models for ordinal phenotypes require improvement for enhanced prediction accuracy.

Purpose of the Study:

• To evaluate and compare the genomic prediction performance of machine learning methods (MLP, SVM) against the TGBLUP model for ordinal phenotypes.
• To assess the accuracy of candidate genotype selection using different statistical models.
• To analyze computational efficiency alongside prediction accuracy.

Main Methods:

• Comparison of Multi Layer Perceptron (MLP), Support Vector Machine (SVM), and Bayesian threshold genomic best linear unbiased prediction (TGBLUP) models.
• Utilized seven real plant breeding datasets with ordinal phenotypes.
• Employed Percentage of Cases Correctly Classified (PCCC) as the primary metric for prediction performance evaluation.

Main Results:

• The TGBLUP model achieved the highest prediction accuracy (PCCC) in four out of seven datasets.
• Support Vector Machine (SVM) demonstrated the least favorable prediction performance but offered the best computational efficiency.
• Multi Layer Perceptron (MLP) models showed competitive accuracy, with no significant difference observed between one, two, or three layers.

Conclusions:

• The TGBLUP model is recommended for improved genomic prediction accuracy of ordinal phenotypes in plant breeding.
• MLP and SVM models provide competitive alternatives, especially when computational time is a constraint.
• Further research into optimizing machine learning models for genomic prediction in plant breeding is warranted.