A two-stage sparse selection method for extracting characteristic genes

Abstract

In this paper, we view gene selection as a feature selection issue, evaluate the response to the abiotic stresses of the genes in a holistic way and propose the Two-Stage Sparse Selection (TSSS) method for gene data analysis. When performing feature selection, our method properly takes into account the interactions and correlations among thousands of genes. Moreover, the two stages of our method have the following rationale: the first stage discards the genes that have very weak relationships with the abiotic stresses and preserves the others that are likely related to the abiotic stresses. The second stage further enables the remaining genes to show their relationships with the abiotic stresses in a competitive way. The experiments show that our method has promising performance in catching the genes that have the strong relationship with the abiotic stresses.