Informative Co-Data Learning for High-Dimensional Horseshoe Regression.

Area of Science:

• Genomics
• Biostatistics
• Computational Biology

Background:

• High-dimensional data are common in clinical genomics for identifying trait predictors.
• Incorporating prior knowledge (co-data) can enhance predictive model performance.

Purpose of the Study:

• To develop a novel Bayesian regression model for high-dimensional data that integrates co-data.
• To improve variable selection and prediction accuracy by leveraging external information.

Main Methods:

• Developed the informative Horseshoe regression (infHS) model.
• Implemented Gibbs sampler for moderate dimensions and Variational approximation for large-scale data.
• Regressed prior variances of regression parameters on co-data variables.

Main Results:

• The simulation study demonstrated the benefits of including co-data.
• The infHS model showed superior performance compared to existing methods in two genomics applications.
• The Variational approximation enabled efficient analysis of very large datasets.

Conclusions:

• The infHS model effectively incorporates co-data into high-dimensional regression.
• This approach offers improved variable selection and predictive performance in genomics.
• infHS provides flexible computational tools for different data scales and inference goals.