Supervised Clustering of Genes for Multi-Class Phenotype Classification

Abstract

The paper presents the new approach to the supervised gene selection by means of gene clustering for the microarray data, which belong to more than two phenotypes (classes). The main distinction from the previous approaches, that are based on the splitting the multi-class task into several binary ones, is the application of the HUM (hypervolume under the manifold) score, that guides the search for the most discriminative gene clusters that simultaneously differentiate all the classes. The results of comparative analysis with other methods shows the advantages of our approach both in classification rate of the new samples and in the lower number of gene clusters. The application of our approach to a randomly permuted data shows that the identified structure is more than just a noise artifact.