Evaluating dimensionality reduction for genomic prediction

Area of Science:

• Plant breeding
• Genomics
• Statistical genetics

Background:

• Genomic selection (GS) utilizes genomic data for early line selection in plant breeding.
• High-dimensional marker data from genotyping presents challenges for statistical modeling in GS.
• Integrating vast genomic datasets into predictive models requires efficient pre-processing techniques.

Purpose of the Study:

• To evaluate the effectiveness of dimensionality reduction (DR) methods as a pre-processing step for genomic selection (GS).
• To compare the performance of five DR methods across different prediction models.
• To analyze the impact of feature reduction on prediction accuracy in GS.

Main Methods:

• Applied five distinct dimensionality reduction (DR) techniques.
• Utilized three statistical models incorporating line, environment, marker effects, and genotype-by-environment interactions.
• Tested methods on a real dataset of 315 lines, 9 environments, and 26,817 markers.

Main Results:

• A small subset of features was sufficient to attain maximum prediction accuracy across DR methods and models.
• Dimensionality reduction significantly enhanced computational efficiency for large genomic datasets.
• The choice of DR method and prediction model influenced the degree of accuracy improvement.

Conclusions:

• Dimensionality reduction methods are valuable pre-processing tools for genomic selection.
• DR enhances computational efficiency in GS by managing large-scale genomic data.
• These findings support the integration of DR into plant breeding pipelines for improved performance.