Testing for significant clustering in dendrograms of gene expression

Question

[4galaxy77]

I am looking for some help on a statistical test I could use on my dendrogram of gene expression. I have used the dendrogram/heatmap function in pheatmap to produce this, to see whether my samples cluster in any particular pattern. Obviously you can visually inspect the figure, but I would like something a bit more rigorous.

For example, it might be that the alpha individuals cluster together more than would be expected by chance (although it clearly looks as though this isn't the case). Can anyone recommend a program (perhaps an R package) which could implement a test of this nature? Thanks.

[ejm32]

Have a look at pvclust in R. It performs a multiscale bootstrap analysis of clusters from hierarchical clustering methods. https://cran.r-project.org/web/packages/pvclust/index.html

[4galaxy77]

Ah that looks perfect, thanks!

[Jean-Karim Heriche]

pvclust gives you an idea of how strongly the data support the clustering but this won't tell you if alphas cluster together more than expected by chance.

[Benn]

Hierarchical clustering is also called unsupervised clustering, hence without prior knowledge. If you want to test if alpha is different from beta, you should use something like limma to find differentially expressed (DE) genes between the groups. If you'll cluster then with the DE genes only you'll probably get better heatmaps.

Answer 1

[Jean-Karim Heriche]

You could try adapting the CLEAN (CLustering Enrichment ANalysis) method. An alternative would be to make disjoint clusters and test whether they are enriched in a particular type compared to the population.

[4galaxy77]

thank you for your help. i will have a look at this.