Fully Adaptive Elastic-net (FAElastic) for gene selection in high dimensional cancer classification

Abstract

Classification of cancer in high dimensional DNA microarray data establish a significant field of research. Though, because of the challenges face by higher dimensional data in selection of genes and classification, numerous penalized likelihood methods are unsuccessful in identifying a small subset of significant genes. To address this problem, the present study proposed and applied a Fully Adaptive Elastic-net (FAElastic) model to perform gene selection and estimation of gene coefficients simultaneously. The proposed techniques, FAElastic-net has been assessed in terms of AUC, number of genes selected, Sensitivity, Specificity and informedness. From the findings which was computed from colon cancer microarray data set, it was confirmed that FAElastic outperforms the other four techniques from the performance metrics which includes: (i) selected number of genes (ii)AUC (iii)Sensitivity and Specificity and (iv) informedness. In addition, FAElastic results can be used practically to other related data of high dimensionality for cancer classification. Thus, we can accomplish the efficiency of the proposed FAElastic-net technique in practice to the medical research area.