An Ensemble Filtering and Supervised Clustering based Informative Gene Selection Algorithm in Microarray Gene Expression Data

Sample classification is one of the important downstream analysis of microarray gene expression data. The outcome of microarray technology is a gene expression data matrix containing expression level of a huge number of genes for a small number of samples which creates difficulty in sample classification. This high-dimensional input space is an overhead for classification algorithms and only a small number of genes contain relevant information for sample classification. In, this regard, a new multi filtering ensemble approach based supervised gene clustering algorithm is proposed here. The proposed method works in two phases. In, the first phase most relevant genes are selected using a multi filtering ensemble approach. In the second phase, the selected relevant genes are clustered using sample class information to find groups of correlated genes closely associated with the sample categories. Then a number of most significant cluster representatives are selected as features for sample classification. The proposed method is compared with different well known existing clustering based and popular filter based gene selection algorithms with respect to the classification accuracy of k-nearest neighbor, Support vector machine and Naive Bayes classifiers on different microarray gene expression datasets. The proposed method significantly outperforms in all cases.