High-dimensional sparse vine copula regression with application to genomic prediction.

Area of Science:

• Statistics
• Genomics
• Machine Learning

Background:

• High-dimensional data in genome-enabled predictions often exhibit complex nonlinear relationships.
• Current vine copula-based regression methods struggle with scalability in high and ultra-high dimensions.
• Efficient statistical modeling is crucial for accurate genomic prediction.

Purpose of the Study:

• To propose novel high-dimensional sparse vine copula-based regression methods.
• To enhance computational efficiency compared to existing approaches.
• To improve variable selection and prediction accuracy in high-dimensional genomic data.

Main Methods:

• Development of two novel high-dimensional sparse vine copula regression techniques.
• Definition of relevant, irrelevant, and redundant explanatory variables for quantile regression.
• Application of methods to simulated and real high-dimensional genomic data for maize trait prediction.

Main Results:

• The proposed methods demonstrate superior computational complexity.
• Effective identification of relevant variables and enhanced prediction accuracy were observed in simulations.
• The methods outperformed linear models and quantile regression forests on real maize genomic data.

Conclusions:

• The novel vine copula regression methods are effective for high-dimensional sparse genomic prediction.
• These methods offer significant advantages in computational efficiency and predictive performance.
• The approach advances the analysis of complex genomic datasets for trait prediction.