A novel neural network approach to gene clustering

Abstract

Clu.stering is a very usefidl and important technique for analyzing gene eJ)pression data. The self organizing nmap has shown to be one of the mlost useful clutstering algorithms. Hlowever, its applicability is limited by the fact that sonme knowledge abotut the clata is required prior to clustering. In this paper we introdutce a novel model of SOM, called growing hierar-chical self-organizing map (GIISOM) to c luster gene expression data. The training and growth process of the GI-ISOM is entirely dcata driven, requiiring no prior knowiledge or estinmates for p)aramneter specification, thtus helps to fintd not only the cappropriate number of cluisters bult also the hiera,'chical relations in the clata set. To validate oulr ressults, we employed a novel validation techniquie, wvhich is k-nown as figure of merit (FOM).