Composite Expectile Regression with Gene-environment Interaction

TL;DR

This paper introduces a novel composite expectile regression method tailored for high-dimensional data with heteroscedastic errors, enhancing the analysis of gene-environment interactions in lung cancer data.

Contribution

It proposes a sparse composite expectile regression model with a coordinate descent algorithm, providing theoretical guarantees and demonstrating superior performance in simulations and real data analysis.

Findings

Method achieves comparable or better performance than existing approaches.

Successfully identifies gene-environment interactions in LUAD data.

Proves selection and estimation consistency of the proposed model.

Abstract

If error distribution has heteroscedasticity, it voliates the assumption of linear regression. Expectile regression is a powerful tool for estimating the conditional expectiles of a response variable in this setting. Since multiple levels of expectile regression modelhas been well studied, we propose composite expectile regression by combining different levels of expectile regression to improve the efficacy. In this paper, we study the sparse composite expectile regression under high dimensional setting. It is realized by implementing a coordinate descent algorithm. We also prove its selection and estimation consistency. Simulations are conducted to demonstrate its performance, which is comparable to or better than the alternatives. We apply the proposed method to analyze Lung adenocarcinoma(LUAD) real data set, investigating the G-E interaction.