Ridge regression in prediction problems: automatic choice of the ridge parameter

Area of Science:

• Genetics
• Statistical genetics
• Bioinformatics

Background:

• Genetic variants are linked to traits, but explain little heritability.
• Standard multiple regression struggles with high-dimensional genetic data.
• Penalized regression methods offer solutions for complex genetic prediction.

Purpose of the Study:

• To develop and validate a data-driven method for choosing the ridge parameter in ridge regression.
• To improve the predictive performance of models using genetic variants.
• To address computational challenges in applying ridge regression to large genetic datasets.

Main Methods:

• Focus on ridge regression, a penalized regression technique.
• Develop a novel method for automatic ridge parameter selection.
• Implement the method in an R package named 'ridge' for efficient computation.

Main Results:

• The proposed method provides theoretical justification for automatic ridge parameter selection.
• Demonstrated good performance on simulated and real genetic data.
• The 'ridge' R package effectively handles millions of genetic variants.

Conclusions:

• The developed method enhances the utility of ridge regression for genetic prediction.
• The 'ridge' R package offers a practical solution for large-scale genetic analyses.
• This approach improves the prediction of phenotypic traits from genetic data.