Supervised Neural Networks for Clustering Conditions in DNA Array Data After Reducing Noise by Clustering Gene Expression Profiles

Abstract

In this paper we compare various applications of supervised and unsupervised neural networks to the analysis of the gene expression profiles produced using DNA microarrays. In particular we are interested in the classification of samples or conditions. We have found that if gene expression profiles are clustered at the optimal level, the classification of conditions obtained using the average gene expression profile of each cluster is better than that obtained directly using all the gene expression profiles. If a supervised method (a back propagation neural network) is used instead of an unsupervised method, the efficiency of the classification of conditions increases. We studied the relative efficiencies of different clustering methods for reducing the dimensionality of the gene expression profile data set and found that the Self-Organising Tree Algorithm (SOTA) is a good choice for this task.