Describir: Use of relevancy and complementary information for discriminatory gene selection from high-dimensional gene expression data.