Hi Guys,

My first question here. I'll make it as short and concise as possible. Please correct me if I am wrong (which I very well might be).

I would like to obtain between sample normalized, within sample normalized gene expression values, e.g. size factor adjusted TPM values. I have quantified my RNA-seq experiment using Salmon and imported the results with tximport to do the differential analysis with DESeq2. Tximport allows the use of the TPM values from Salmon to do the differential analysis, but entails a transformation of said values to either scaledTPM or lengthScaledTPM values.

As I understand it, the scaledTPM values are the between-sample normalized TPM values multiplied by the library size in millions. So something like "transcripts pr sample", which is not really what I want.

My question is, would it be possible to somehow output between sample normalized TPM values for each gene or is this somehow violation a principle I am overlooking?

Could I just divide the the TPMs from Salmon with the sizeFactors obtained from DESeq2?

Cheers,

Michael

You might consider dividing out the size factors returned by estimateSizeFactorsForMatrix on the TPMs, but not from the counts. I've done this before, in the alpine paper for example. The reason one might consider to re-normalize TPM (in theory library size has been removed) for some datasets or some tasks is that it is possible for very highly abundant genes with small errors in measurement to induce errors in other less abundant transcripts. So you end up, using the technique above, with something like TPM, but which reduces this problem by assuming that the median TPM ratio between samples should be 1. It will not sum to 1e6 exactly for samples though, but that's not a problem in my opinion.