Figure 1.
Sample pairwise similarity matrix for fuzzy partition matrices’ rows relabeling.
Each element in this sample pairwise matrix measures the similarity between a cluster from one clustering result and a cluster from another clustering result. In the min-max relabelling approach, the minimum value of each column is calculated, as shown in the row below the matrix, and then the maximum of these minima is considered. The maximum of the minima is shaded in dark gray and the clusters corresponding to the row and the column containing this value are matched. This row and this column are then removed and the process is repeated until each cluster in the first result is matched with a cluster from the second result.
Table 1.
Clustering experiments.
Figure 2.
The number of correctly assigned genes at the y-axis is plotted versus the 16 binarization configurations at the x-axis for three representative synthetic datasets out of 60. It should be noted that the binarization configurations are not entirely ordered according to their tightness.
Figure 3.
False-positives index (FPI) is plotted in log scale versus a subset of binarization configurations for three representative synthetic datasets out of 60. It should be noted that the binarization configurations are not entirely ordered according to their tightness.
Figure 4.
False-negatives index (FNI) is plotted in log scale versus a subset of binarization configurations for three representative synthetic datasets out of 60. It should be noted that the binarization configurations are not entirely ordered according to their tightness.
Figure 5.
SNR effect over the number of multiply assigned and unassigned genes.
(a) The number of multi-assigned genes is plotted over the 60 SNR values in four cases of wide clusters generated by using the TB technique. (b) The number of unassigned genes is plotted over the 60 SNR values in four cases of tight clusters generated by using the DTB technique. Note that there are no multi-assigned genes in tight clusters as there are no unassigned genes in wide clusters.
Table 2.
Budding yeast cell-cycle microarray datasets.
Table 3.
Assignment of genes from five real yeast datasets by Bi-CoPaM.
Table 4.
Comparison between Bi-CoPaM and many existing clustering methods.
Figure 6.
YAL040C / CLN3 gene expression profiles in the five microarray datasets.
Genetic expression profiles for the cyclin CLN3 from the five datasets cdc28, cdc15, alpha, alpha-30 and alpha-38 are plotted. Although the gene’ is known to be expressed periodically, different levels of periodicity for its profiles can be seen for different datasets clearly.
Table 5.
Fuzzy membership values for the CLN3 gene.