A novel neural network approach to gene clustering

Clustering is a very useful and important technique for analyzing gene expression data. The self-organizing map has shown to he one of the most useful clustering algorithms. However, its applicability, is limited by the fact that some knowledge about the data is required prior to clustering. In this paper we introduce a novel model of SOM, called growing hierarchical self-organizing map (GHSOM) to cluster gene expression data. The training and growth process of the GHSOM is entirely data driven, requiring no prior knowledge or estimates for parameter specification, thus helps to find not only the appropriate number of clusters but also the hierarchical relations in the data set. To validate our results, we employed a novel validation technique, which is known as figure of merit (FOM).