A Novel Clustering Algorithm with Map Energy Minimization

Abstract

Cluster analysisisthe fundamentalmethod for DNA microarray gene classi cation[1], structure activ-ity relationship [2] and any kind of correlation matrix analyses. The hierarchical clustering analysis isthe most widely used method providing us clear-separated clusters, however, often distributes closelyrelated clusters into apart positions because the direction of branches at each node is arbitrary. Thislimitation often makes the correlation matrix look like checkered pattern.In this work, we propose a new clustering method to overcome this problem. Our new methodsearches a set of row and column orders in which any pair of the highly coherent rows (or columns)are placed adjacently in the correlation matrix map. The nal map is no longer expected to havecheckered pattern. We assumed that it could perform this process to minimize the energy of thecorrelation map pattern. We named our new method \Map Energy Minimization Clustering (Meminclustering).