Abstract: This paper presents a new method of similarity measurement, on the basis of which an global K-means algorithm is presented that is selfadaptive to the determination of the number of cluster, so as to solve the problem that the existing methods of co-regulated gene clustering can take into account all the four co-regulating relationships, that is, reverse, asynchronous, parttime, and differences in the expression. The present algorithm has been applied to the microarray data, and the compare experimental results with CLUSTER 3.0 prove that the algorithm is efficient.

Key words: co-regulating gene, clustering, global K-means, similarity measurement