What is a good RNA seq normalization method that allows for across sample comparisons and between transcripts

Question

What is a good RNA seq normalization method that allows for across sample comparisons, and allows between transcripts comparisons as well? I read that TMM for example allows across sample comparisons but not between transcripts. On the other hand, I read that TPM is not good enough for between sample comparisons

Answer 1

I'm having a bit of trouble understanding the question, but we've used DESeq2 for doing both between-sample and between-transcript comparisons.

When comparing transcripts for short reads I use the variant-stabilised transformation, adjusted for transcript length. This is good for a visual overview of results (e.g. in a heatmap).

When comparing samples I use the variance-adjusted DESeq2 test, producing normalised $\textrm{log}_2FC$ values. This works well for quantitative results.

See this paper, where we use both of these approaches (e.g. Figure 2 shows both in a side-by-side view).

Answer 2

As mentioned by @gringer DESeq2 is a popular choice for normalization of counts between samples. However, because each gene is adjusted individually it is not well suited to compare within a sample.
For this a global normalization like TMM from edgeR would be more appropriate.

You can find more information in this tutorial from the hbc at Harvard.

Answer 3

I found a recent technique that allows for both, called geTMM.