Leveraging external information by guided adaptive shrinkage to improve variable selection in high-dimensional...

Area of Science:

• Statistics
• Bioinformatics
• Machine Learning

Background:

• High-dimensional data with low sample sizes presents significant variable selection challenges.
• External information, termed 'co-data', can improve variable selection accuracy.
• Co-data, such as variable groupings or prior p-values, is abundant in genomics.

Purpose of the Study:

• To review guided adaptive shrinkage methods that utilize co-data for improved variable selection.
• To discuss the technical aspects and applicability of co-data in prediction models.
• To compare guided shrinkage with other methods like sparse group-lasso.

Main Methods:

• Review of guided adaptive shrinkage methods.
• Adaptation of shrinkage parameters using co-data.
• Comparison with sparse group-lasso for variable selection.
• Integration of co-data learners and spike-and-slab priors for 'do-it-yourself' implementation.

Main Results:

• Guided adaptive shrinkage methods effectively use co-data to enhance variable selection.
• The methodology demonstrates versatility in integrating different co-data types.
• Demonstration of improved variable selection in genetics studies through DIY implementation.

Conclusions:

• Guided adaptive shrinkage offers a powerful framework for variable selection with co-data.
• The methods are applicable across various domains, especially genomics.
• Practical implementation guidance is provided for researchers.