Method for Feature Selection Based on Inference of Gene Regulatory Networks

In this paper, we propose a wrapper method for relevant gene subset-selection based on inference of gene-regulatory networks. This method can solve two important tasks in genomic data analysis. First, it can infer regulatory-networks from gene-expression data using extremely random trees and concepts of network-centrality, second, it can identify a subset of relevant genes for the classification task by dropping regulator genes. We evaluated the proposed method with 6 cancer microarray-gene expression datasets. Datasets present binary and multiclass tasks. For all six datasets, we have inferred the gene-regulatory networks and performed the feature-selection. We trained 4 classifiers using the selected genes and obtained excellent classification performance. Comparison of the proposed method with existing feature selection methods shows that it performs very well.