Among those genes with an adjusted q-value (based on FDR) of < 0.05, authors used hierarchical clustering based on the HOPACH algorithm, which provides a criterion for choosing the optimal number of clusters for each level of the hierarchy by utilizing the silhouette statistic.