Microarray data mining: A novel optimization-based approach to uncover biologically coherent structures

BMC Bioinformatics

Table 3 Comparison of biological coherence of clusters obtained for dataset II by different clustering algorithms

	Proportion of Genes (%) in Clusters of p-values
	<= 10^-4	<= 10^-5	<= 10^-4	<= 10^-5
	(Uncorrected)		(Bonferroni-Corrected)
EP_GOS_Clust	79.4*	66.1*	57.2	52.2*
Iterated EP_GOS_Clust	86.2*	83.9*	73.1*	68.9*
K-Means	78.0	62.1	58.0*	51.7
K-Correlation	77.1	63.9	57.3*	51.8
K-Medians	78.8*	65.3	56.9	52.2*
SOTA	75.2	66.9*	57.1	39.3
IClust	66.0	54.0	34.2	29.1
	Cluster Correlation			-log₁₀(P) Values
	Max.	Min.	Ave.	Average
EP_GOS_Clust	0.920	0.454*	0.730*	9.17*
Iterated EP_GOS_Clust	0.956*	0.489*	0.750*	11.09*
K-Means	0.961*	0.049	0.668	9.01
K-Correlation	0.964*	0.398*	0.717*	9.13
K-Medians	0.923	0.203	0.683	9.09
SOTA	0.911	0.285	0.624	9.20*
IClust	N.A.	N.A.	N.A.	9.01

Methods include EP_GOS_Clust backbone, the iterative algorithm described in this report, the K-family of partitional clustering algorithms with pre-assigned clusters, self organizing tree algorithm (SOTA), and mutual information based clustering (IClust) [28]. Data in the upper table are presented as described in the legend to Table 2 while the lower table presents data on expression correlation within clusters and the average -log(P) values for biological coherence over all the clusters.
* The top three performers in each category are indicated with an asterisk.

ISSN: 1471-2105