Table 2 |
||||||
|
Cluster accuracy and stability on the completely synthetic data with four repeated measurements at low noise level |
||||||
|
(a) Cluster accuracy*
|
||||||
| Algorithm |
Similarity measure/model |
Average |
SD-weighted |
CV-weighted |
FITSS |
IMM |
|
|
||||||
| Average linkage |
Elliptical |
NA |
NA |
NA |
NA |
0.957 (0.089) |
| Complete linkage |
Elliptical |
NA |
NA |
NA |
NA |
0.957 (0.090) |
| MCLUST-HC |
NA |
0.930 (0.100) |
NA |
NA |
0.902 (0.123) |
NA |
| Average linkage |
Distance |
0.877 (0.191) |
0.927 (0.100) |
0.876 (0.192) |
0.804 (0.155) |
NA |
| Centroid linkage |
Distance |
0.840 (0.178) |
0.927 (0.100) |
0.927 (0.100) |
0.840 (0.178) |
NA |
| k-means |
Distance |
0.877 (0.191) |
0.927 (0.100) |
0.876 (0.192) |
NA |
NA |
| Average linkage |
Spherical |
NA |
NA |
NA |
NA |
0.926 (0.102) |
| Complete linkage |
Distance |
0.925 (0.102) |
0.876 (0.192) |
0.925 (0.103) |
0.840 (0.179) |
NA |
| Complete linkage |
Spherical |
NA |
NA |
NA |
NA |
0.897 (0.088) |
| Centroid linkage |
Correlation |
0.765 (0.115) |
0.575 (0.179) |
0.527 (0.141) |
0.765 (0.115) |
NA |
| Average linkage |
Correlation |
0.764 (0.114) |
0.576 (0.201) |
0.597 (0.210) |
0.718 (0.222) |
NA |
| k-means |
Correlation |
0.764 (0.114) |
0.707 (0.140) |
0.652 (0.116) |
NA |
NA |
| Complete linkage |
Correlation |
0.755 (0.116) |
0.584 (0.189) |
0.599 (0.194) |
0.609 (0.201) |
NA |
|
|
||||||
|
(b) Cluster stability
† |
||||||
|
|
||||||
| Algorithm |
Similarity measure/model |
Average |
SD-weighted |
CV-weighted |
FITSS |
IMM |
|
|
||||||
| Average linkage |
Distance |
0.970 (0.055) |
0.998 (0.003) |
0.998 (0.002) |
0.979 (0.018) |
NA |
| Centroid linkage |
Distance |
0.984 (0.025) |
0.998 (0.002) |
0.986 (0.027) |
0.984 (0.025) |
NA |
| k-means |
Distance |
0.954 (0.062) |
0.998 (0.003) |
0.967 (0.070) |
NA |
NA |
| Average linkage |
Correlation |
0.958 (0.054) |
0.832 (0.095) |
0.854 (0.060) |
0.984 (0.025) |
NA |
| Complete linkage |
Distance |
0.968 (0.041) |
0.923 (0.131) |
0.981 (0.037) |
0.918 (0.055) |
NA |
| Average linkage |
Elliptical |
NA |
NA |
NA |
NA |
0.961 (0.081) |
| Complete linkage |
Elliptical |
NA |
NA |
NA |
NA |
0.960 (0.083) |
| Centroid linkage |
Correlation |
0.959 (0.054) |
0.861 (0.119) |
0.866 (0.026) |
0.959 (0.054) |
NA |
| k-means |
Correlation |
0.958 (0.054) |
0.693 (0.144) |
0.634 (0.098) |
NA |
NA |
| Complete linkage |
Correlation |
0.947 (0.067) |
0.580 (0.140) |
0.671 (0.100) |
0.915 (0.049) |
NA |
| MCLUST-HC |
NA |
0.935 (0.089) |
NA |
NA |
0.916 (0.066) |
NA |
| Average linkage |
Spherical |
NA |
NA |
NA |
NA |
0.852 (0.089) |
| Complete linkage |
Spherical |
NA |
NA |
NA |
NA |
0.779 (0.091) |
|
|
||||||
|
*Each entry shows the average adjusted Rand index of the corresponding clustering approach with the six classes. We ran our experiments on five randomly generated synthetic datasets, and show the average results with the SD of the adjusted Rand index in brackets. A high average adjusted Rand index represents close agreement with the classes on average. †Each entry shows the average adjusted Rand index of the original clustering result with clusters from remeasured data. The SD of the average adjusted Rand indices over five sets of randomly generated synthetic remeasured data is shown in brackets. A high average adjusted Rand index means that clusters from remeasured data are in close agreement with clusters from the original data. The external knowledge is not used in evaluating cluster stability. For both parts of the table, the maximum average adjusted Rand index of each row is shown in bold. The algorithms (rows) are sorted in descending order of the maximum average adjusted Rand index in each row. |
||||||
|
Yeung et al. Genome Biology 2003 4:R34 doi:10.1186/gb-2003-4-5-r34 |
||||||
