clustering from breast cancer RNA-seq data

Question

[silviajserrano]

Hello!! I work in breast cancer and I have RNA-seq data from 60 patients. I classified the samples into intrinsic subtypes by immunohistochemistry and I have been using DEseq2 to perform differentially expression analysis between subtypes. I would like to do a global analysis. I mean, I would like to see which samples cluster together according to the gene expression, something unsupervised but using the whole transcriptome. I already used PAM50 to assign samples into intrinsic subtypes according to gene expression but I would like to see what samples cluster together and what characterized each cluster and from there I would like to do the differentially expression analysis…what tools can I use?

Answer 1

[Michael Love]

Have you seen the clustering examples in the DESeq2 vignette or workflow?

vignette("DESeq2")

For 60 samples, I would apply the varianceStabilizingTransformation (see vignette), then explore with the plots we show in the vignette. You can use built in R tools to cluster, for example:

mat <- assay(vsd)
hc <- hclust(dist(t(mat)))
plot(hc)
cutree(hc, k=3)

If you get the latest version of R (3.3) and the latest version of DESeq2 (1.12) you can use an even faster version of VST I just implemented:

vsd <- vst(dds)

Answer 2

[silviajserrano]

Thank you!! and after clustering, How can I explore the gene expression of each cluster? I mean, what characterize each cluster...

[TriS]

How To Get The Subclusters From The Object Of Hclust() Using Cutree() According To The Order On The Map Produced By Heatmap.2? this is a way you can do it

Answer 3

[skjobs]

For the RNA seq or higher dimensional matrix, I will recommend you to use the tSNE algorithm for unsupervised clustering i.e. PCA.