Adaptive group-regularized logistic elastic net regression

Area of Science:

• Bioinformatics
• Statistical Genomics
• Computational Biology

Background:

• High-dimensional omics data analysis often benefits from incorporating external feature information.
• Leveraging prior knowledge such as p-values or annotations can improve classification and feature selection.
• Integrating this external information into statistical models presents analytical challenges.

Purpose of the Study:

• To develop a novel group-regularized logistic elastic net regression method (gren) for high-dimensional data.
• To effectively incorporate external feature information into omics data analysis.
• To enhance classification performance and feature selection accuracy.

Main Methods:

• Proposed a group-regularized logistic elastic net regression method (gren).
• Utilized a Bayesian formulation of logistic elastic net regression.
• Employed an approximate empirical-variational Bayes framework for parameter estimation.
• Applied the method to cancer genomics and Alzheimer's metabolomics datasets.

Main Results:

• The gren method successfully integrated external feature information.
• Demonstrated enhanced classification performance compared to standard methods.
• Showcased improved feature selection accuracy when external information was informative.
• Validated findings through simulations and real-world omics studies.

Conclusions:

• The proposed gren method effectively enhances classification and feature selection in high-dimensional omics data.
• Incorporating informative external feature data through group regularization is beneficial.
• gren provides a robust framework for leveraging prior biological knowledge in omics analysis.