Interacting gene selection via cooperative game analysis for cancer diagnosis.

Abstract

Microarray technologies offer practical diagnostic tools for cancer detection. One great challenge is to identify salient genes from the high dimensionality of microarray data that can directly contribute to the symptom of cancer. Interactions among genes have been recognized to be fundamentally important for understanding biological function. This paper proposes an interacting gene selection method for cancer classification by identifying useful interacting genes. The method firstly evaluates the interactivity degree of each gene according to the intricate interrelation among genes by cooperative game analysis. Then genes are selected in a forward way by considering both interactivity and relevance characters. Experimental comparisons are carried out on four publicly available microarray data sets with three outstanding gene selection methods. Moreover a gene set enrichment analysis is also performed on the selected gene subset. The results show that the proposed method achieves better classification performance and enrichment score than other gene selection methods.