Lasso-penalized Cox model for compositional mutational signatures via the L1-CoDa norm

I. Galván Femenía, J. Saperas Riera, J. Palarea Albaladejo

Mutational signatures represent the imprints of biological processes accumulated during tumoral evolution. Expressed as proportions, these signatures reside in the simplex. Despite their prognostic value, current Cox proportional hazards models often treat them as independent covariates or dichotomize them, ignoring their constrained nature. We propose a Lasso-penalized Cox framework adapted for the simplex by extending the L1-CoDa regularization to survival analysis. Originally formalized for linear regression (Saperas-Riera et al., 2023), we adapt this methodology to the Cox log-partial likelihood. By treating the estimation as a constrained regularization task, we ensure identifiability and scale-invariance within the Aitchison geometry. Using data from 1,069 colorectal cancer patients, we demonstrate that this bridge between penalized likelihood and compositional geometry offers a robust framework for survival modeling in high-dimensional oncology.

Keywords: Mutational signatures, compositional data analysis, Cox proportional hazards, L1-CoDa regularization

Biostatistics II
September 4, 2026  3:30 PM
Aula 24

P. Jurado Bascón, J. A. Villatoro García, P. Carmona Sáez

Ó. Lado-Baleato, S. Pérez-Fernández

B. Monteiro, A. H. Tavares, T. Gregório

B. Piñeiro Lamas, A. López-Cheda, R. Cao