Coxmos: interpretable survival models for high-dimensional and multi-omic data.

Area of Science:

• Biostatistics
• Bioinformatics
• Computational Biology

Background:

• High-dimensional (HD) and multi-block (MB) survival analysis, common in omics, faces challenges like multicollinearity and low interpretability.
• Traditional Cox models struggle with HD data, while machine learning methods often lack transparency.
• Existing Partial Least Squares (PLS) survival models have limited tools for optimization, evaluation, and interpretation.

Purpose of the Study:

• Introduce Coxmos, an R package designed for robust and interpretable survival analysis in HD and MB settings.
• Provide integrated tools for variable selection, model optimization, validation, comparison, interpretation, and visualization.
• Address limitations of existing methods in omic and multi-omic survival data analysis.

Main Methods:

• Developed Coxmos, an R package integrating adapted Cox regression with HD-variable selection.
• Implemented optimized PLS-based approaches specifically for HD and MB data.
• Incorporated validation, comparison, interpretation, and visualization functionalities within the package.

Main Results:

• Coxmos demonstrated superior performance compared to state-of-the-art machine learning methods on diverse cancer datasets (BRCA, HNSC, HGSOC).
• The package significantly enhances model interpretability in complex survival analysis.
• Case studies on HGSOC and HNSC illustrated Coxmos's utility in model selection, validation, and biological interpretation.

Conclusions:

• Coxmos offers a flexible, robust, and interpretable solution for survival analysis in HD and MB data.
• The package facilitates reliable survival prediction and identification of risk/protective factors in complex biomedical research.
• Coxmos fills a methodological gap, providing a valuable resource for reproducible and interpretable survival modeling in big data scenarios.