Advice on organizing large GSVA heatmap

Question

[Morgan S.]

Hi there,

I wanted to get some advice on how you might make your heatmap easier to read. In my case, I generated a heatmap from GSVA data which I filtered to only include significant pathways, here.

I wanted to see how each of my samples differed in the expression of these pathways. In the heatmap, there are 130 significant pathways, so my heatmap is extremely large and would be impossible to use as a figure in a manuscript.

If you had this figure, how would you try to simplify it? Would you find some way to cluster the GO terms into related pathways and generate multi heatmaps of these different pathway types?

As you can see, there are two well-separated large clusters as a determined heatmap, but I do not see a clear reason in this separation.

Anywho, I look forward to your opinions and suggestions :)

Answer 1

[LauferVA]

Hey Morgan,

Might help to check out arguments that can be input to the R package pheatmap at CRAN to get some ideas.

Specifically, if you go to the pheatmap man page, and navigate to the second page of the pdf, you will see arguments like "cluster rows" or "cluster columns".

Depending on what you want to visualize, you could group the rows or columns by anything you wanted.. for isntance you could put all cases and all controls next to each other.

Alternatively, you could group them by how well the data group together in the first place ... irrespective of case control (or other) covariate status. If, for example, you found that all the males group together very very well, and one female groups with them, this might be a reason to go back and make sure that female is correctly labelled.

But depending on what you are trying to use the heatmap to show, clustering cols may or may not be helpful. You can get additional ideas by scanning the other arguments for pheatmap in that pdf, or by looking at examples... e.g. pheatmap vignettes.

[Morgan S.]

Hi Vincent,

What I ended up doing was looking at the enriched KEGG pathways instead of the individual GO terms to simplify the analysis. Then plotted with pheatmap and clustered the rows (pathways).

Thanks! :)