Hello,

I was wondering if someone might be able to explain why there are sometimes overlapping colors module color section of plots created by WGCNA plotDendroAndColors?

For example, in the manual network construction tutorial (2. Network construction and module detection -> b. Step-by-step network construction and module detection) on the WGCNA website, there is a lot of striation of the module color assignments, particularly on the left-side of the plot.

Intuitively, I would expect the color assignments (which correspond to branch cuts in the dendrogram above) to be sequential, and not mixed together in this manner -- e.g. all of the "magenta" module would be in one place, and so on.

Any ideas why the color assignments appear this way?

First, remember that the order of the objects on a clustering tree is to some degree arbitrary. Each merge can be flipped, and the algorithm that R uses for choosing which of the two possible flips to use is designed to result in visually informative clustering trees, but does not order objects using any kind of dissimilarity information.

To understand why parts of clusters apparently overlap (at least according to the dendrogram), look at Figure 3 of the Supplement http://labs.genetics.ucla.edu/horvath/CoexpressionNetwork/BranchCutting/Supplement.pdf or at this plot:

Look at the Simulated turquoise and red branches. They merge at a relatively low height and there are extra singletons just above the merge that should be assigned to one of the two branches (clusters). But the dendrogram itself provides no information about which branch each singleton is closest to. That's why the "hybrid" dynamic tree cut contains the second Partitioning Around Medoids (PAM)-like stage in which each object is assigned to its nearest cluster. This can lead to alternating colors above the merge of two clusters and simply reflects the fact that when an object sits above the merge of two branches, the dendrogram provides no information about which of the two branches the object is closest to. In this sense, the alternating colors are a reflection of a limitation of the clustering tree as a measure of object proximity.

Hope this helps,

Peter

Hi Keith,

See here  http://www.genetics.ucla.edu/labs/horvath/CoexpressionNetwork/BranchCutting for more information.

Best,

Jim