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

Motivation: Survival analysis is fundamental in many fields, including medicine. However, the analysis of high-dimensional datasets, such as omic or multi-omic data, presents significant challenges due to data dimensionality, multicollinearity, and the limited interpretability of existing methods. Classical Cox models struggle in these contexts, and machine learning approaches often lack transparency. Results: To overcome these limitations, we present Coxmos, an R package that integrates adapted Cox regression models with variable selection and PLS-based approaches tailored for high-dimensional and multi-block data. Coxmos also provides validation, comparison, interpretation, and visualization tools. Benchmarking across clinical, omic, and multi-omic datasets showed that Coxmos outperformed other state-of-the-art machine learning methods, while offering biological interpretation. We showcased Coxmos functionalities on an ovarian cancer dataset, highlighting its potential to integrate multiple omic layers, identify relevant predictors, and assess their impact on patient survival.