Skip to main content
. 2014 Feb 14;30(12):1774–1776. doi: 10.1093/bioinformatics/btu098

Table 1.

Mean TM-score of centroids relative to native structure

Clustering methoda Centroid of largest clusterb Best centroid of five largest clustersb
Spicker 0.584 0.607
Clusco/k-means/RMSD 0.585 0.612
Multi-k-means/RMSD 0.590 0.624
Multi-k-means/TM-score 0.592 0.624
Hierarchical/RMSD 0.588 0.626
Hierarchical/TM-score 0.595 0.624

afast_protein_cluster k-means values are the average of five separate runs to control for different starting seeds. Distance matrices were calculated using CA-atom coordinates. bTM-score means that are significantly better (paired z-test with P < 0.05) than Spicker are in bold, and those significantly better than Clusco are underlined. The quality of the best model among the centroids of the five largest clusters is significantly improved when fast_protein_cluster is used as the clustering method.