Normalisation of RNA-seq data for coexpression analysis

Question

[DanTzo]

Hello everyone,

I've been wondering on what is the best, or most widely accepted, way to normalise RNA-seq data (either FastQ or feature counts) from multiple samples, in order to get a gene expression dataset that is best for global coexpression analysis. I have tried using qsmooth and vst normalisation but I have not been satisfied with the results.

Pointings to any existing literature on this subject would be also much appreciated.

Thanks everyone in advance.

[MatthewP]

Have you tried TPM?

[DanTzo]

Thank you for your reply Matthew,

Initially I tried using TPM, however, the resulting grouping wasn't as good as I expected. That is why I tried using read counts normalised with the methods I mentioned, although in retrospect, TPM might have resulted in a better product.

Is there evidence that TPM is better in general for this kind of analysis?

Answer 1

[Kevin Blighe]

I have tried using qsmooth and vst normalisation but I have not been satisfied with the results.

In which way were you not satisfied?

As the term implies, a co-expression analysis is based on correlation; so, any 'cleaned' and normalised data can be tolerated.

My own suggestion would have been VST, possibly followed by an additional transformation to Z-scores.

Kevin