A regularized Cox hierarchical model for incorporating annotation information in predictive omic studies

Area of Science:

• Bioinformatics
• Statistical Genomics
• Computational Biology

Background:

• High-dimensional omics data often includes informative meta-features like biological pathways and functional annotations.
• These meta-features can enhance prediction of outcomes, particularly time-to-event data.

Purpose of the Study:

• To introduce a regularized hierarchical framework for integrating meta-features with omics data.
• To improve prediction and feature selection performance for time-to-event outcomes.

Main Methods:

• A hierarchical framework was developed to incorporate meta-features.
• Regularization was applied at both omics and meta-feature levels to handle high-dimensional data.
• The model was fitted using iterative reweighted least squares and cyclic coordinate descent.

Main Results:

• The regularized hierarchical model substantially improved prediction performance compared to standard Cox regression when meta-features were informative.
• Applications to breast cancer and melanoma survival data demonstrated improved prediction and identified important omics feature sets.
• The model showed robustness, performing comparably to standard methods when meta-features were uninformative.

Conclusions:

• The hierarchical regularized regression model effectively integrates external meta-feature information for time-to-event outcomes.
• It enhances prediction accuracy and aids in discovering important features, serving both predictive and discovery applications.
• The framework is robust, maintaining performance even with uninformative external data.