DegreeCox - a network-based regularization method for survival analysis

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• High-dimensional molecular data poses challenges for oncological survival modeling.
• Traditional methods like LASSO (Least Absolute Shrinkage and Selection Operator) offer interpretability but don't fully utilize feature relationships.
• Graph-based feature representations can capture complex biological networks.

Purpose of the Study:

• To introduce DEGREECOX, a method applying network-based regularizers for Cox proportional hazard models.
• To utilize gene network centrality measures to constrain models for patient survival prediction.
• To address the challenge of high dimensionality in oncological data analysis.

Main Methods:

• DEGREECOX infers Cox proportional hazard models using network-based regularizers.
• Network centrality measures (weighted degree, betweenness, closeness) are employed.
• A priori network information is sourced from Gene Co-Expression Networks and Gene Functional Maps.

Main Results:

• DEGREECOX demonstrated improved classification of high and low-risk ovarian cancer patients compared to RIDGE and LASSO.
• Performance was comparable to NET-COX, particularly for challenging, less separable datasets.
• Root Mean Square Error (RMSE) and Concordance Index (C-index) showed competitive results, slightly exceeding top methods in some instances.

Conclusions:

• Network-based regularization is a promising strategy for high-dimensional survival data.
• Centrality metrics can be extended to incorporate diverse biological network topological properties.
• This framework offers a robust approach to enhance patient survival predictions in oncology.