Highly correlated modules with clinical trait does not cluster together in eigengene dendrogram, WGCNA.

Question

[vipulwagh31]

I am trying to identify clinically relevant hub genes using WGCNA. To do so, I have followed WGCNA tutorial, 'Network analysis of liver expression data from female mice'. I have used the step-by-step method and signed network to generate modules. However, I am confused about the way modules cluster together in the eigengene dendrogram. For example, the module lightgreen is highly correlated with the trait (image 1) but does not cluster together with that of the clinical trait in the eigengene dendrogram (image 2). As far as I understand, the module-trait relationship and eigengene dendrogram use eigengene values to establish the correlation with any given trait. If that's the case, I would expect lightgreen module clustered with that of the clinical trait in the dendrogram as well.

Similarly, the modules greenyellow and salmon which are clustered together with the trait, are not strongly correlated with the trait as seen in the module-trait relationships heatmap.

Below are the links for the images.

1. Image for module-trait relationship:

2. Image link for Eigengene dendrogram

3. Image link for Eigengene adjacency heatmap

4. Scale independence and mean connectivity

5. Number of genes assigned to the modules

I would really appreciate any explanation for this. I am confused and I am really not sure how do I choose the modules?

Thanks in advance

[andres.firrincieli]

I would expect lightgreen module clustered with that of the clinical trait in the dendrogram as well

They do not cluster together because the sign of the correlation is negative (Image 1).

the modules greenyellow and salmon which are clustered together with the trait

while they share the same branch they do not cluster together. Just look at the distance between the nodes (Image 3)

[vipulwagh31]

@andres.firrincieli: Thanks for your quick response.

They do not cluster together because the sign of the correlation is negative (Image 1).

So, does that mean that the positively correlated modules will always cluster together with that of trait in the eigengene dendrogram and the negatively correlated ones will be placed far in the dendrogram?

while they share the same branch they do not cluster together. Just look at the distance between the nodes (Image 3). Can you please elaborate on this a bit. I have no idea how to look at the nodal distance in the adjacency heatmap. I am sorry for that.

[andres.firrincieli]

So, does that mean that the positively correlated modules will always cluster together with that of trait in the eigengene dendrogram and the negatively correlated ones will be placed far in the dendrogram?

Exactly. For example, if you look at the Eigengene adjacency heatmap positively correlated modules tend to cluster together in the Eigengene dendrogram (i.e. black, magenta and midnightblue).

Can you please elaborate on this a bit. I have no idea how to look at the nodal distance in the adjacency heatmap. I am sorry for that.

Sorry, I was referring to Image 2. This is how you read a dendrogram. I can't find a better explanation.

Enjoy

[vipulwagh31]

Thanks alot for your response again.

I'd surely go through the literature recommended.