Extraction of diverse gene groups with individual relationship from gene co-expression networks

TL;DR

This paper presents a novel method combining graphical embedding and recursive partitioning to extract diverse, biologically plausible gene groups from co-expression networks, especially from clinical data, improving the understanding of gene relationships.

Contribution

It introduces a new module detection approach optimized for clinical gene co-expression data, enhancing diversity and biological relevance of identified gene groups.

Findings

High stability and biological plausibility of gene modules

Effective in analyzing cancer-related gene expression data

Improved assessment of gene relationships in human data

Abstract

Motivation: Modules in gene coexpression networks (GCN) can be regarded as gene groups with individual relationships. No studies have optimized module detection methods to extract diverse gene groups from GCN, especially for data from clinical specimens. Results: Here, we optimized the flow from transcriptome data to gene modules, aiming to cover diverse gene relationships. We found the prediction accuracy of relationships in benchmark networks of non-mammalian was not always suitable for evaluating gene relationships of human and employed network based metrics. We also proposed a module detection method involving a combination of graphical embedding and recursive partitioning, and confirmed its stable and high performance in biological plausibility of gene groupings. Analysis of differentially ex-pressed genes of several reported cancers using the extracted modules successfully added relational information consistent with previous reports, confirming the usefulness of our framework.