This thesis focuses on analyzing high-dimensional microarray data using the proposed algorithm, Variational-SOM. The original Self-Organizing Map (SOM) algorithm is an unsupervised neural network method and can be used to reduce the dimensionality of microarray data. The main disadvantage of SOM is that the topology of the map must be fixed from the beginning. In order to solve the problem, the Variational-SOM, of which the map's topology is determined dynamically, is proposed.The DNA microarray technology makes it possible to monitor expression levels of thousands of genes simultaneously. However, these data are of little use unless we are able to analyze them.Experimental results show that the Variational-SOM can reduce the dimensionality of data according to the information that the data contains and help to extract biological significance from the data. The analysis using Variational-SOM can produce more well-separated clusters with respect to clinical information than using the original SOM.