Table 2. Performance of the clustering algorithms across all simulated data and all Broad Institute TCGA data.
| Hierarchical | K-means | Divisive hierarchical | Self-organizing maps | Self-organizing trees | Partitioning around medioids | Clustering for large applications (CLARA) | Agglomerative Nesting | |
|---|---|---|---|---|---|---|---|---|
| A. Simulated Data | ||||||||
| K3 No Noise | 0 | 0 | 1.9 | 5.6 | 0 | 0 | 13.7 | 0 |
| K3 Noise 1000 | 0 | 0 | 5.7 | 5.6 | 0 | 0 | 6.1 | 0 |
| K5 No Noise | 0 | 0 | 7.4 | 5.6 | 0 | 0 | 16.3 | 0 |
| K5 Noise 1000 | 0 | 0 | 5.2 | 5.6 | 0 | 0 | 14.6 | 0 |
| K7 No Noise | 0 | 0 | 9.3 | 5.6 | 0 | 0 | 14.8 | 0 |
| K7 Noise 1000 | 0 | 0 | 0.6 | 5.6 | 0 | 0 | 20 | 0 |
| B. Broad Institute TCGA Data | ||||||||
| BRCA | 0 | 0 | 20.2 | 1 | 2.4 | 0 | 12.5 | 0 |
| COADREAD | 0 | 0 | 37 | 2.2 | 12.3 | 1.3 | 19.4 | 0 |
| GBM | 0 | 0 | 15 | 0 | 18.7 | 0 | 1.3 | 0 |
| KIRC | 7.1 | 6.9 | 35.9 | 6.9 | 22.2 | 15.8 | 20.7 | 10.4 |
| LUSC | 0.3 | 0.3 | 20.2 | 1.3 | 10.4 | 1.3 | 16.2 | 1.5 |
| OV | 0 | 0 | 22.6 | 1.2 | 0.2 | 0 | 11.6 | 0 |
| UCEC | 4.9 | 2.5 | 32.5 | 5.1 | 16.3 | 22.7 | 21.7 | 9.3 |
Shown is the percent of runs for which each algorithm produced any clusters with three or fewer clusters, across all subsets.