A Novel Clustering Approach: Global Optimum Search with Enhanced Positioning - Clustering Challenges in Biological Network - page 318

Fig. 16.1. Schematic flowchart of the EP GOS Clust algorithm. Although the formulation in the
paper has been given for DNA microarray data, the algorithm framework can be adapted for clustering
any numeric data.
16.3. Results and Discussion
16.3.1. Description of Comparative Study
We will work with the 5652 genes obtained previously. The clustering algorithms
to be compared are (a) K-Means, (b) K-Medians, (c) K-Corr, where the Pearson
correlation coefficient is the distance metric, (d) K-CityBlock, where the distance
metric is the city block distance, or the 'Manhattan' metric, which is akin to the
north-south or east-west walking distance in a place like New York's Manhattan
district, (e) K-AvePair, where the cluster metric is the average pair-wise distance
between members in each cluster, (f) QTClustering, (g) SOM, (h) SOTA, (i) GOS
I, where genes with up to 7 different feature points are pre-clustered, initial clus-
ters are defined by uniquely-placed genes, and each gene is placed into its nearest
cluster as the initialization point, and (j) EP GOS Clust, for which genes are pre-
clustered if they have 2 or less different feature points and can be uniquely clus-