Performance determinants of unsupervised clustering methods for microbiome data