We used VXOrd to layout the metagene co-expression network. We used the negative logarithm of the P-values of the co-expression of each pair of metagenes as the similarity measure given to the force-directed placement procedure of VXOrd. The layout procedure situated each metagene in the X-Y plane
Identification of components.
We visually inspected the landscape produced by VXOrd and identified 12 regions of highly interconnected meta-genes that we refer to as components. We then used a K-means clustering procedure to identify which genes were included in each of the 12 components. Using the X-Y coordinates of the meta-genes as determined by VXOrd, we seeded the K-means procedure with 12 centers approximately equal to the locations of the 12 components. The procedure converged in 3 iterations. Meta-genes that were farther away than d units from their closest center were excluded from membership in any of the 12 components. We chose d to be 10% of the diameter of the entire landscape.