This paper shows an iterative clustering method based on kernel k-means, which changes the parameter k automatically in each iteration of the algorithm. In addition, a way to initialize the centroids is proposed. The method is applied to a binning process in metagenomics using a complex database with different organisms. The aim of this method is to reduce the sensitivity of clusters based on strength measures. The results demonstrate that the proposed method is better than the simple kernel k-means for metagenome databases.