A tool for cluster number estimation in SOM-based gene expression pattern análisis

FERANDEZ, ELMER ANDRES; BALZARINI, MÓNICA

Angra dos reis, Brazil

Simposio; Brazilian Bioinformatics Symposium; 2007

Abstract. Cluster methods are crucial to study genomic patterns of coexpressed genes. Neural network algorithms such as self organizing map (SOM) have been extensively used to cluster gene expression data. Result visualization techniques are important tools for cluster recognition in SOM. In this work a simple tool that implements an algorithm to identify and visualize clusters is proposed. It is based on two concepts (Relative Position and Q statistics) that can be applied to a SOM network. The Relative Position is a new SOM node-adaptive attribute defined from the node moving within a two dimensional space imitating the movement of the SOM codebook vectors in the input space. By means of the Q statistics the algorithm evaluates the SOM structure providing an estimate of the number of clusters underlying the data. The tool allows the visualization of the cluster node patterns facilitating cluster interpretation.