Regulation-incorporated Gene Expression Network-based Heterogeneity Analysis

TL;DR

This paper introduces a novel high-dimensional penalized fusion method for heterogeneity analysis in gene expression networks that accounts for regulatory effects, improving the identification of gene network structures and heterogeneity in cancer data.

Contribution

It proposes a new heterogeneity analysis method based on gene expression networks that incorporates regulation effects and uses a penalized fusion approach with proven consistency properties.

Findings

The method effectively identifies heterogeneity and gene network structures.

It outperforms existing methods in simulations.

Application to breast cancer data reveals biologically meaningful heterogeneity.

Abstract

Gene expression-based heterogeneity analysis has been extensively conducted. In recent studies, it has been shown that network-based analysis, which takes a system perspective and accommodates the interconnections among genes, can be more informative than that based on simpler statistics. Gene expressions are highly regulated. Incorporating regulations in analysis can better delineate the "sources" of gene expression effects. Although conditional network analysis can somewhat serve this purpose, it does render enough attention to the regulation relationships. In this article, significantly advancing from the existing heterogeneity analyses based only on gene expression networks, conditional gene expression network analyses, and regression-based heterogeneity analyses, we propose heterogeneity analysis based on gene expression networks (after accounting for or "removing" regulation effects) as well as regulations of gene expressions. A high-dimensional penalized fusion approach is proposed, which can determine the number of sample groups and parameter values in a single step. An effective computational algorithm is proposed. It is rigorously proved that the proposed approach enjoys the estimation, selection, and grouping consistency properties. Extensive simulations demonstrate its practical superiority over closely related alternatives. In the analysis of two breast cancer datasets, the proposed approach identifies heterogeneity and gene network structures different from the alternatives and with sound biological implications.