Hierarchical clustering to group large number of samples

Question

[nazaninhoseinkhan]

Dear all,

I want to cluster around 250 normalized htseq count samples to get insight about how many distinct groups are present in my samples.

I used hierarchical clustering using "Euclidean distance measure" available in GENE-E program.

However I don't know is it OK if I select visually separated groups? At which level (distance) can I define separate groups?

Any comments is appreciated

Nazanin

Answer 1

[Jean-Karim Heriche]

It is perfectly reasonable to cut a hierarchical clustering tree based on visual inspection. A fixed level or distance threshold is often inadequate. If you want an automatic way of cutting the tree, you can try the R package dynamicTreeCut.
Alternatively, once you know from hierarchical clustering what kind of structure there is in your data, you could try other clustering algorithms that could automatically produce the desired clusters.

[nazaninhoseinkhan]

Thanks for your comments.

Sure. I will try dynamicTreeCut package