Improving Genomic Prediction Using High-Dimensional Secondary Phenotypes: The Genetic Latent Factor Approach.

Area of Science:

• Plant breeding and genetics
• Genomics
• Bioinformatics

Background:

• Advancements in high-throughput phenotyping (HTP) generate large, high-dimensional datasets.
• Integrating HTP data into genomic prediction faces challenges like multicollinearity and computational complexity.
• Existing methods often struggle with parameter interpretability.

Purpose of the Study:

• To develop a novel method for integrating secondary HTP data into genomic prediction.
• To address challenges associated with high-dimensional data in plant breeding.
• To improve the accuracy and interpretability of genomic prediction models.

Main Methods:

• Proposed genetic latent factor best linear unbiased prediction (glfBLUP) pipeline.
• Utilized generative factor analysis to reduce HTP data dimensionality.
• Estimated genetic latent factor scores using filtered and regularized correlation matrices.
• Applied latent factors in a multitrait genomic prediction framework.

Main Results:

• glfBLUP demonstrated superior performance compared to alternative methods in simulations and a real-world case.
• The method effectively reduces data dimensionality while maintaining predictive power.
• Generated interpretable and biologically relevant model parameters.

Conclusions:

• glfBLUP offers a flexible and modular framework for multitrait genomic prediction.
• The approach enhances genomic prediction accuracy by leveraging HTP data.
• Provides a foundation for more interpretable and powerful genomic selection strategies in plant breeding.