Hierarchical clustering was done by using the clustering analysis implemented in the Genomic Workbench software.
In addition, 2D descriptor based clustering was done to differentiate the ligands, binding to closed, semi-open and open conformations of human AK.
Thus the clustering was done in a qualitative manner, relying on researchers' judgments.
The 40 clusters obtained by k-means clustering were globally common divisions, because the clustering was done using all polygon attributes.
Clustering was done using Euclidian distance method after normalization (shift probe set expression to mean of zero and scale to standard deviation of 1).
Clustering was done by means of a standard single-linkage algorithm with a Euclidean metric (Matlab, Mathworks).
Clustering was done with the Cluster program [48] using Pearson correlation as similarity metrics and centroid linkage clustering.
Clustering was done with the public available MALTAB implemented Ncutclustering_7 toolbox of Shi (http://www.cis.upenn.edu/jshi/software).upenn.edu/jshi/software
Non-hierarchical Affinity-Propagation clustering was done using the apcluster.m algorithm [30] with the preference value p = 0.7.
Hierarchical clustering was done on centered genes by using the centered correlation distance and an average linkage method.
Visualization plots and hierarchical clustering was done using SpotFire software version 8.2.1 with package DecisionSite for Functional Genomics.
