Gene ontology driven feature selection from microarray gene expression data

One of the main challenges in the classification of microarray gene expression data is the small sample size compared with the large number of genes, so feature selection is an essential step to remove genes not relevant to class label. Traditional gene selection methods often select the top-ranked genes based on their individual discriminative powers. The problem with these simple ranking models is that they evaluate genes in isolation and this may introduce redundancy among the selected feature subset. Most redundancy based methods solely evaluate gene expression levels. This may decrease the effectiveness of feature selection since some values may not be accurately measured. In this paper, we propose a gene ontology based method for feature selection. The novelty of this model is to detect redundancy between a pair of genes by the convex combination of their expression similarity and semantic similarity in gene ontology. The effectiveness of our method is demonstrated by the experiment in two widely used datasets.