I have been working on clustering using RNAseq data. To compute distance, what kinds of normalization is optimal?

Can we use normalization using relative log normalization? I understand this should be used for differential gene expression analysis, not for unsupervised learning in which we do not use information of disease condition.

If you can show good papers, please share.

  • 1
    $\begingroup$ Hi it would be useful to describe the disease in question and some idea of the data. "Normalisation" is often used in parametric statistics to convert data to a normal distribution as a pre-requisite of the analysis, or linear modelling. Transformation is a "hit and miss" business in general. Unsupervised learning ... it depends what you mean, but you wouldn't do this for PCA or tSNE. More details please. $\endgroup$
    – M__
    Commented May 21, 2020 at 6:19

1 Answer 1


Typically some sort of variance stabilizing transform is used before clustering. Popular options are regularized log transformation or a vst transform, which are available in DESeq2. Note that these are NOT used for performing differential expression, just things like clustering and PCA. For differential expression one would use TMM, RLE or similar.

  • $\begingroup$ Thank you for your comment. Yes, I would like to calculate the distance to perform clustering, not differential expression. $\endgroup$
    – user224050
    Commented May 21, 2020 at 14:56
  • $\begingroup$ So if we want to clustering after calculating distance, you mean we should use regularized log transformation or a vst transform. And when we want to perform differential gene expression analysis, we should use TMM, RLE. So we need to change normalization methods in one paper depending on analytic plan, right?? $\endgroup$
    – user224050
    Commented May 21, 2020 at 15:01
  • $\begingroup$ Yes, though regularized log and such are transforms and not normalizations. $\endgroup$
    – Devon Ryan
    Commented May 21, 2020 at 15:10
  • $\begingroup$ So for the analytic pipeline to compute the distance, we should not use both normalization and transformation ,right? raw count (without transformation)→vst or other transformation. towardsdatascience.com/… $\endgroup$
    – user224050
    Commented May 22, 2020 at 16:01
  • $\begingroup$ Please see and follow the DEseq2 tutorial, this is covered there. $\endgroup$
    – Devon Ryan
    Commented May 22, 2020 at 19:40

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.