Penalized composite quantile regression model for compositional data

doi:10.1080/02331888.2026.2632740

Penalized composite quantile regression model for compositional data

2026-02-01

0

OA

AI

DOI：10.1080/02331888.2026.2632740

Original

Original request for help

Save

Abstract

En 中文

Motivated by challenges in gut microbiome data analysis, we propose a penalized composite quantile regression (CQR) method with linear constraints that ensures non-crossing quantiles across multiple levels. The method is particularly well suited for compositional data, while the framework naturally extends to more general constrained penalized CQR settings. By integrating information across multiple quantiles, the proposed approach improves estimation efficiency, especially in the tails, while preserving robustness without requiring the errors to follow any particular distribution. To solve the associated optimization problem, we develop an efficient ADMM algorithm. Simulation studies and a microbiome application investigating the relationship between gut composition and human body mass index illustrate the robustness, accuracy, and practical utility of the method.

Keywords:

Composite quantile regression

penalized and constrained

non-crossing quantiles

compositional data

ADMM algorithm

Penalized composite quantile regression model for compositional data

Abstract

Journal

Organization