A New Model of Identifying Differentially Expressed Genes via Weighted Network Analysis Based on Dimensionality Reduction Method

Abstract

Background: As a method to identify Differentially Expressed Genes (DEGs), Non- Negative Matrix Factorization (NMF) has been widely praised in bioinformatics. Although NMF can make DEGs to be easily identified, it cannot provide more associated information for these DEGs. Objective: The methods of network analysis can be used to analyze the correlation of genes, but they caused more data redundancy and great complexity in gene association analysis of high dimensions. Dimensionality reduction is worth considering in this condition. Methods: In this paper, we provide a new framework by combining the merits of two: NMF is applied to select DEGs for dimensionality reduction, and then Weighted Gene Co-Expression Network Analysis (WGCNA) is introduced to cluster on DEGs into similar function modules. The combination of NMF and WGCNA as a novel model accomplishes the analysis of DEGs for cholangiocarcinoma (CHOL). Results: Some hub genes from DEGs are highlighted in the co-expression network. Candidate pathways and genes are also discovered in the most relevant module of CHOL. Conclusion: The experiments indicate that our framework is effective and the works also provide some useful clues to the reaches of CHOL.