Unbiased clustering from bulk tissue RNA-seq?

Question

[gspirito]

Hello everyone, I have a question:

I have bulk tissue whole RNA-seq data from patients and controls. I would like to know whether I could find different sub-groups of patients based solely on the transcriptional landscape, so no pre-defined sample groups. I performed a simple k-mean clustering analysis with R's function kmeans() (both on raw counts and on TPM), by imposing an arbitrary number of clusters. The results could make sense, but I would like to compare these results with some coming from a method more specific to RNA-seq data.

Could anyone point me to some tool?

Thanks in advance,

Giovanni

Answer 1

[Kevin Blighe]

The results could make sense, but I would like to compare these results with some coming from a method more specific to RNA-seq data.

k-means is fine to use for the TPM expression levels. Do not worry.

Please take a look at the ComplexHeatmap examples, where a lot of this can be made easier for you:

• 2.3 Clustering

Kevin

Answer 2

[swbarnes2]

People use PCA for this as well, to eyeball how the samples group. Check out a deseq tutorial to see how to do that right.

Answer 3

[ATpoint]

Very recently published is SDCM, an unsupervised clustering technique for expression data that you can check. The talk I heard was impressive since they managed to find a very recently published subgroup of DLBCL with it but without specifically tuning parameters for it. Afaik it contains a link to the GitHub repository. https://www.nature.com/articles/s41467-019-12713-5